ABSTRACT. Purpose The role of image-guidance in tumor ablation procedures is at least twofold; to guide the physician to the target and to provide an assessment of ablation margins. We report our experience in having performed 125 PET/CT guided abdominal tumor ablations in the PET/CT scanner in the Advanced Multimodality Image-Guided Operating (AMIGO).

Methods Patients with primary or metastatic tumors that are not ideal surgical candidates are treated in the AMIGO PET/CT suite. Tumors of the liver, kidney, lung, adrenal, peritoneum, lymph nodes and other structures are treated with various ablation technologies including microwave, cryo, radiofrequency and chemical ablation or more recently with irreversible electroporation. PET/CT cases are often performed under general anesthesia or sometimes with monitored anesthesia care (moderate to deep IV sedation). Advantages of FDG PET tumor visibility include persistent tumor visibility throughout the procedure. Metallic devices which produce CT artifacts do not obscure PET visibility of tumor. Edema or hemorrhage may obscure the tumor on CT images, yet the metabolic tumor activity is unchanged on PET. While image-fusion technologies have been an intense focus of research and product development for interventional applications, they are limited by the many factors related to fusing retrospective data sets, whether obtained weeks earlier or minutes earlier in the procedure. PET/CT overcomes most of these obstacles through near real-time acquisition of both the PET and CT scans at any time-point of the procedure. No matter what transpires during an interventional procedure (device insertions, creation of artificial ascites, hemorrhage, tissue contraction from microwave ablation, etc.) acquiring both PET and CT scans in a single breath-hold ensure near-ideal image fusion of the current anatomical configuration. Tumor FDG activity is not dissipated by thermal ablation; therefore, tumors remain visible as metabolically active masses after ablation. With contrast-enhanced CT or US, the tumor is typically obscured within the ablation zone. This makes assessment of the ablation margin, and in particular the minimum ablation margin, very difficult. Fusion to retrospective data sets do not factor in anatomical distortions of the procedure including tissue contraction. PET/CT offers three ways to depict the entire tumor ablation margin through tumor visibility on PET and ablation zone depiction on CT.

Conclusion We have leveraged the many advantages that PET/CT offers to develop successful strategies for interventional oncology, including depicting the entire tumor ablation margin intraprocedurally. These could potentially guide overlapping ablations before concluding the procedure.

Funding Source(s): Siemens Healthineers Grant, NIH P41EB015898 (AMIGO)

ABSTRACT. Endovascular Ion Exchange ChemoFilter Reduces Off-Target Doxorubicin Exposure in a Hepatic Intra-Arterial Chemotherapy Model Colin Yee1, Jay Yu1, Aaron Losey, MD1, Caroline Jordan1, Mark Wilson1, and Steven Hetts1 1 University of California, San Francisco, San Francisco, CA Purpose: Current locoregional therapies for treating primary liver cancer, such as transarterial chemembolization (TACE) and intraarterial chemotherapy (IAC), selectively treat isolated tumors by injecting chemotherapeutics into the arteries supplying the tumor. Even though TACE and IAC have been shown clinically to increase local tissue concentrations and reduce systemic exposures, off-target toxicities continue to be the main deterrent towards effective treatments. Doxorubicin (DOX) is a highly potent antineoplastic drug used to treat a wide spectrum of cancer types, but due to its off-target toxicities the administered doses are restricted to cumulative life time dose of 360-500 mg/kg2. Therefore, we evaluated the ability of endovascular chemofiltration using novel ionic devices to reduce systemic exposure and off-target biodistribution of DOX during hepatic (IAC) in a preclinical model. Methods and Materials: The devices were percutaneously introduced via the internal jugular vein and deployed in the right hepatic vein and near the confluence spanning the IVC to the right atrium. Consecutive infusions of 200 mg of DOX (2 mg/ml) were injected over 10 min into the common hepatic artery feeding the liver. Treatment animals underwent the first infusion without devices in place, but during the second dose chemofilters were deployed. While control animals were administered both infusions without devices. Peripheral blood samples and tissue DOX concentrations were measured to evaluate the devices abilities reducing systemic exposure and drug accumulation. Results: Animals treated with CF devices showed a consistent reduction in peak DOX concentrations compared to their baseline DOX infusions, decreasing on average 32.0% ± 7.2%, while in controls the peak DOX [C] increased over the consecutive infusions on average 7% ± 3.5%. Treatment with endovascular CF devices showed dramatic 82% reduction in DOX concentrations in the heart from 26.4 ± 2.5 (μg/g) in controls without devices to 14.3 ± 2.3 (μg/g) in animals treated with chemofilters. There was found to be a significant position dependence affecting device DOX adsorption. CF devices placed near the hepatic confluence with proximal shafts traversing the suprahepatic IVC bound significantly more DOX compared to devices deployed in the infrahepatic IVC. The DOX distinctly fluoresced verifying the CF device bound drug in situ (Fig. 1). Conclusion: Image-guided endovascular chemofiltration with ionic endovascular devices placed non-occlusively in the hepatic venous outflow and inferior vena cava during hepatic artery doxorubicin infusion reduced off-target blood and tissue concentrations of doxorubicin in a translational model.

ABSTRACT. Studies suggested tissue autofluorescence can be used to generate contrast images between different tissues for cancer detection. Current fluorescence lifetime imaging systems capture the emission decaying profile with fast accurate sampling, and the data is then fitted to an exponential decaying model to compute the decay times. The noticeable postprocessing time required to extract the decay times and point measurement method limit its intraoperative application. Our group developed a real-time imaging technique using a fast-gated charge-coupled device (CCD) camera and pulsed diode array. In stead of capturing the complete decaying profile of a single point, we aim to generate a wide field of view contrast image by developing spatially resolved maps of relative differences in auto-fluorescence decay of tissue constituents. Our developed algorithm utilizes two images taken during excited and decaying states, both subtracted by a background image, to generate a map proportional to the relative decaying fluorescence lifetime. The relatively short acquisition time and simple algorithm utilized in our system significantly decreased the overall time needed to generate an image and make the intraoperative real-time imaging possible.

ABSTRACT. Purpose: The trans-oral approach to head and neck cancer resections has decreased surgical morbidity due to its minimally-invasive nature. Intraoperative assessment of tumor extent and localization of key anatomical structures beneath the visible mucosal surface, however, remain challenging because of the anatomical distortion caused by retraction and laryngoscope placement. A custom-designed CT/MR-compatible laryngoscopy system developed in our lab enables intraoperative imaging, intraoperative navigation, and evaluation of intraoperative tissue deformation. Pressure sensors incorporated into a laryngoscope provide real-time force measurements during laryngoscopy. Fusion of all these data streams is being used to develop a predictive deformation model that updates intraoperatively to provide more accurate surgical guidance during trans-oral surgery (TOS).

Methods: A custom CT/MR-compatible polymer laryngoscopy system was designed to provide equivalent functionality to standard metal laryngoscopes (Fig A). Preoperative and intraoperative CT scans were performed on fifteen patients in order to quantitatively evaluate tissue deformation (bony structure displacement, tongue movement, and airway volume change) associated with suspension laryngoscopy. Intraoperative navigation was trialed on four patients during TOS using a tracked probe and Medtronic StealthStation S7 System. A custom force sensing laryngoscopy system (Fig B) was deployed on cadavers and in one patient.

Results: Intraoperative CT scans with our system showed clear improvement in image quality as compared to those acquired with traditional metal laryngoscopes (Fig C), enabling segmentation of key anatomical structures and reliable navigation (Fig D). In the subset of 4 patients in which instrument tracking was trialed, registration accuracy of 1.5mm or less was achieved. Intraoperative deformation of up to 4cm and airway volume change of up to 108.9cm3 were noted in the patient cohort. Upwards of 60lbs of force were recorded on the scope surface.

Conclusions: Our MR/CT-compatible laryngoscopy system allows for intraoperative imaging and navigation to visualize and quantify anatomical deformations due to instrumentation. Real-time force measurement during laryngoscopy is clinically feasible using instrumented laryngoscopes. Intraoperative imaging and force sensing provide data streams necessary for designing and validating biomechanical tissue deformation models and registration algorithms. Future work will use our instrumented laryngoscope’s position and force information to deform biomechanical airway models and validate them through intraoperative imaging.

Funding Source(s): This work was supported by Dartmouth’s Clinical and Translational Science Institute (UL1TR001086) from NIH’s National Center for Advancing Translational Sciences and by Dartmouth-Hitchcock Norris Cotton Cancer Center’s Prouty Pilot grant and Hopeman Fund for Clinical Research.

ABSTRACT. Purpose: To develop an automated platform that defines determinate habitat volumes in the prostate that may then be more reproducibly targeted for radiotherapy dose escalation, and that facilitates the development of tools for assessing early response quantitatively. Background: Decision-making in prostate cancer, from whether to biopsy, to assessment of risk through accuracy in tumor sampling, to treatment selection and management choices, is dependent on the identification and characterization of at risk volumes in 3D. While the use of multiparametric MRI (mpMRI) has gained momentum because of improvements in diagnostic reliability, there remains considerable variability in interpretation and, until now, no automated, quantitative, reproducible, way of defining such volumes. We describe a pixel by pixel 10 level habitat risk scoring (HRS) system that has been incorporated into a targeted biopsy and radiotherapy workflow. Materials and Methods: HRS was optimized by referencing to prostatectomy Gleason score in 3D with rigid fusion to pretreatment mpMRI Dynamic Contrast Enhanced MRI (DCE-MRI) and Apparent Diffusion Coefficient (ADC) sequences. A workflow for RT planning was created where the HRS contours are migrated to the planning CT using a randomized Phase II mpMRI targeted boost clinical trial (BLaStM; NCT02307058) as the platform. Results: There were 51 regions of interest (ROIs) in 12 patients who underwent radical prostatectomy (RP). The resultant heat maps showed inter- and intra-tumoral heterogeneity. The HRS6 level was significantly associated with RP ROIs (slope 1.09, r=0.767; p<.0001). For predicting the likelihood of cancer, GS≥7 and GS≥8, HRS6 AUCs were 0.718, 0.802 and 0.897, respectively. HRS was superior to the PIRADS 4/5 classification (difference with HR6, p<.0001). HRS maps were created for the first 37 assessable patients on the BLaStM trial. There were an average of 1.38 habitat boost volumes per patient at a total boost volume average of 3.6 cc. Conclusions: An automated quantitative mpMRI based method was developed to objectively guide dose escalation to high risk habitat volumes based on prostatectomy GLSC. Funding Source(s): This work was supported by National Institutes of Health grants R01CA189295 and R01CA190105.

ABSTRACT. Colonoscopy is an invasive procedure with several limitations including patient discomfort/use of sedation, procedural access, and long learning curve. To overcome these limitations, our team has developed a magnetic flexible endoscope (MFE), actuated using an external permanent magnet (EPM) mounted on a robotic arm, with the pose of the capsule sensed in real time by a localization system. Using this pose information, we demonstrate that image based lumen detection using the capsules on board camera can semi-autonomously guide the capsule to reach the cecum of a human colon phantom.

ABSTRACT. Purpose: The superior ability of MRI to image soft tissue malignancies has led to an increase in its usage during needle-based percutaneous procedures such as biopsy, cancer brachytherapy, and tumor ablation. A technical challenge associated with this interventional use of MRI is the difficulty in interpreting the artifacts created by the metallic needles that are placed into suspicious or cancerous tissue. We are leveraging advances in deep artificial neural networks, or deep learning, that have revolutionized computer aided diagnosis and detection to aide interventional oncology by developing automatic needle segmentation and visualization methods for prostate biopsy and gynecologic cancer brachytherapy.     

Methods: We have optimized a fully convolutional neural network (FCN) architecture for automatic segmentation of needles from T2-weighted MRI images, and applied it (under IRB approved studies) to two separate retrospective databases containing nearly 600 biopsy needles and nearly 800 high dose brachytherapy needles in pelvic MRI, respectively.  Needles were annotated on each MRI, which were acquired intra-procedurally at our institution. The patients were divided in the ratio 70:30 into two independent training-validation and test sets. A deep 3D FCN model was designed, trained and deployed on these samples. An asymmetric architecture with in-plane pooling and up-sampling layers was used to account for anisotropy in the MRI volumes. The accuracy of the developed architecture, as tested on previously unseen data, was 2.8 mm average in needle tip detection, which is well within the clinically acceptable range. An observer study in which independent annotations by a second rater were compared to the output of the proposed method, and the resultant error was comparable to the measured inter-rater concordance, reinforcing the clinical acceptability of the proposed method.  

Conclusion: MRI is increasingly being used in interventional oncology procedures to guide percutaneous needle insertions for soft tissue biopsies, tumor ablations, and brachytherapy.  While it is the not ideal imaging modality for visualizing metallic needles, we have developed promising deep neural network based post-processing methods that can be used for automatic segmentation and hence visualization of needles for interventional oncology.

Funding Source: NIH P41EB015898

ABSTRACT. Purpose

Twin-to-twin transfusion syndrome is a potentially fatal disease of placental vasculature that can affect pregnancies involving twin fetuses that share a single placenta. It is treated with fetoscopic laser photocoagulation surgery, a procedure in which a surgeon uses a fetal endoscope to selectively cauterize abnormal vascular connections on the placenta. This procedure is made difficult by the poor visibility conditions of the uterine environment, which can partially or completely obscure blood vessels. Given the difficulty of this surgery, there is value in creating educational tools for trainee surgeons, including surgical simulators.

Methods

A cyclic generative adversarial network (CycleGAN) [1] is a form of neural network that is trained on two sets of images A and B and learns to synthesize B images from A images. Endoscopic videos were obtained from fetoscopic surgeries performed at Yale-New Haven Hospital in a process approved by an IRB. The videos were downscaled and cropped from 1920 × 1080 pixels to 256 × 256 pixels (Fig 1a). Next, 700 schematic diagrams of fetoscopic video frames were generated programmatically. These diagrams had RGB color channels in which blue denoted the field of view, green denoted blood vessels, and red denoted glare (Fig 1b). A CycleGAN was trained to synthesize endoscopic video frames from schematic diagrams. The training dataset was assembled by randomly pairing still frames from the endoscopic video dataset with schematic diagrams. The network was trained for 200 epochs by gradient descent.

Results

The synthetic video frames produced by the CycleGAN bear a striking resemblance to real video frames while retaining the structural elements of the schematic diagrams.

Conclusions

It is possible to generate a realistic simulation of intraoperative video from a schematic diagram of placental vasculature. This could be valuable in creating training simulations for surgeons. Additionally, when paired with the schematics, the synthetic frames could be used as training data for fetal endoscopic video segmentation algorithms.

Funding Source(s)

This work was supported by the National Institutes of Health grant number T35DK104689 (NIDDK Medical Student Research Fellowship).

References

[1] J.-Y. Zhu et al., “Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks.” ICCV, 2017.

ABSTRACT. Purpose: Primary liver cancer and liver metastases are leading causes of worldwide cancer morbidity and mortality. Surgery results in five year survival rates of 30-60% in selected patients, however, less than 20% of patients are surgical candidates. As long-term survival is only possible with the addition of local therapies to systemic therapies, there is strong rationale for improved and increased use of local therapies in primary and metastatic liver cancer, including radiation therapy and focal ablative therapy. There is strong evidence to suggest that advances in image guidance can improve targeting, reduce toxicities, and improve the understanding of these focal therapies. The purpose of this work is to improve the integration of advanced imaging for the focal treatment of liver cancer. Methods: A biomechanical model-based deformable image registration (DIR) algorithm, Morfeus, has used to investigate improvements in targeting liver tumors for radiotherapy and focal thermal ablation, as well as modeling the response of the liver to both focal therapies to improve our understanding of the response of normal liver. To improve the efficiency of Morfeus, which relies on segmentation of the liver, an auto-segmentation algorithm was developed using a fully-convolutional neural network. To improve the targeting of the tumor for focal microwave ablation, a retrospective analysis was performed using Morfeus to map the tumor, defined on the diagnostic CT image, onto the ablation image and subsequently onto the post-ablation image where the ablation region was identified. The ablation margin was quantified after Morfeus mapped the tumor onto the ablation region to assess the ablation margin for 14 patients who had local tumor progression and 16 patients who had no evidence of local tumor progression (mean follow-up 28.8 months). In addition, Morfeus was used in combination with an automated segmentation of the vasculature within the liver to perform DIR between radiotherapy planning images and images post-radiotherapy where registration is challenging due to complex volume changes from hypertrophy and fibrosis. Nineteen patients were retrospectively evaluated using vessel bifurcations identified on the corresponding images to calculate the target registration error (TRE). 3D vessel segmentation was performed using tubular structure enhancement and Otsu thresholding. A centerline extraction was performed and a distance map was calculated to find corresponding points to be used as boundary conditions. The results were compared to rigid registration, Demon’s algorithm, and standard Morfeus without the internal boundary conditions. Results: The auto-segmentation of the liver was preferred to the clinician-drawn contours in 52% of the cases, 92% of the auto-segmentation results were deemed clinically useable. Using Morfeus to map the tumor onto the ablation region demonstrated a statistically significant difference in the minimum distance to agreement from the tumor to the ablation region between the patients who had a local recurrence and the patients who did not. Patients who did not have a local recurrence had on average twice the ablation margin of those who recurred, 3.2 mm compared to 1.1 mm. Using Morfeus with vascular-based boundary conditions to register the pre-radiotherapy liver to the post-radiotherapy liver resulted in a TRE of 5.3 mm (SD 3.1mm) compared to rigid (8.9 mm, SD 2.9 mm), Demon’s (12.6 mm SD 7.4 mm), and Morfeus without internal boundary conditions (9.4 mm, SD 3.4 mm). Conclusions: This work demonstrates the multi-disciplinary application of a biomechanical DIR to improve targeting and the correlation of the treatment-based images with follow-up images. This work was funded in part by NIH 1R01CA221971.

ABSTRACT. Interventional image-guided therapies (IGT) typically benefit from real-time feedback about patient anatomy and tool placement. Interventional magnetic resonance imaging (iMRI) is particularly challenging for navigation. We describe a novel, clinical, vision-based image guidance system that allows for real-time navigation of arbitrary needle-like tools using fused iMRI and ultrasound (US) imaging. We have established an iMRI workflow at a top-5 hospital’s IR department. Combined MRI and US fusion imaging can now be used intraoperatively instead of repeated iMRI scans. We report performance and stability numbers.

ABSTRACT. Purpose: Treatment for gynecologic malignancies may include high-dose-rate (HDR) interstitial brachytherapy (ISBT) to increase the dose delivered to the tumor relative to nearby healthy tissues. This procedure requires a temporary implant with multiple hollow needles (typically about 12 needles) inserted through a perineal template. Despite the need for precise needle placement to avoid overexposing organs at risk (OAR), including the bladder, rectum, and bowel, and deliver optimal treatment, there is no standard method for intraoperative visualization or localization of needles. Implementation of an accessible intraoperative system for verifying needle positions would facilitate image guidance for the implant, potentially avoiding OAR and improving implant quality. The purpose of this study was to design a 3D ultrasound (US)-based guidance system and assess its use for visualizing needles during gynecologic ISBT implant procedures. Methods: We developed a 3D US system that can be used for either transrectal US (TRUS) or transvaginal US (TVUS) imaging, depending on the patient’s anatomical and tumour geometry. This system uses a motorized mechanism to rotate a conventional 2D side-fire endocavity US probe through 170° for TRUS acquisition to create a fan-shaped image or 360° for TVUS acquisition to create a ring-shaped image (Figure 1), reconstructing the 2D frames into a 3D image available immediately after acquisition (12-20 s acquisition time). Additionally, during 3D TVUS scanning, the template vaginal cylinder is replaced with a hollow, sonolucent cylinder to accommodate the probe. The template is registered to the TVUS image and the clinician has the option to update subsectors of the image or for real-time 2D US visualization in the context of the 3D image and help to guide the needle during insertion. Images were acquired for five patients undergoing HDR ISBT for each mode of the system, with 58 and 46 needles localized in the TRUS and TVUS images, respectively. For each patient, the post-insertion computed tomography (CT) and corresponding 3D US images were rigidly registered and for each needle identified, the needle tip or visible point closest to the tip (exit point) was selected and one other point on the needle path for comparison between modalities. Results: The mean needle tip difference in the 3D TRUS images was 3.82 ± 1.86 mm with an angular difference of 3.04 ± 1.63°. The mean exit point difference in the 3D TVUS images was 1.91 ± 0.81 mm and the angular difference was 1.85 ± 0.68°. In addition to the needles, OAR, including the bladder (with Foley catheter) and rectum were clearly visible in the 3D TVUS images. Conclusions: The 3D US-based system allowed needles to be visualized and localized during HDR gynecologic ISBT, using two different acquisition modes. TRUS was particularly useful for visualizing posterior needles and tips extending beyond the template cylinder, while TVUS allowed visualization of needle trajectories in all areas of the template and nearby OAR. This provides the potential for 3D image guidance during implant insertion procedures. Funding Sources: Ontario Institute of Cancer Research & Canadian Institutes of Health Research. JR was supported by the Natural Sciences and Engineering Research Council, Cancer Research and Technology Transfer Program, & Ontario Graduate Scholar Program.

ABSTRACT. PURPOSE: In this work we report on a novel framework to test a real-time deformation-corrected image guided liver navigation system in open liver surgery. To our knowledge, this is the first ever real-time evaluation of intraoperative deformation-corrected image guided liver surgery. METHODS: We introduce a novel bystander approach for assessing the real-time value of deformation-corrected image-guided liver surgery. Briefly described, during open surgery an optically tracked stylus was use to digitize the anterior liver surface data. Both a rigid, and a model-based deformable registration is performed using this sparse localization data. The novel real-time evaluation entails evaluating approximately 6-7 guidance displays in succession for each patient, all being assessed with the patient on the OR table. We should note that the displays have been designed such that there is no visual cue within the imaging data itself that allows for discerning rigid from deformable registrations. With each successive guidance display evaluation, the surgeon is asked on a +3 to -3 scale whether the display is better or worse than the previous one evaluated where 0, +1, +2, +3 represent no improvement, small, moderate, and highly improved, respectively (analogous negative scores would reflect worsening). The assessment technique is the procedural standard of interrogating the registration by comparing display output while using the stylus on the organ surface. To enhance scientific rigor, the sequence order of the display transitions is randomized and blinded to the surgeon and system technician. The study reported involved two different surgeons over n=20 cases (n=13, and n=7 cases for each surgeon respectively). Over the n=20 cases, there were 55 instances of the rigid registration display being followed by the deformable registration display, and 46 instances in the reverse. There were 24 instances of registration displays being held constant. RESULTS: Table 1 summarizes the results and reflects a consistent average/median rating that indicates a perceived improved fidelity when transitioning from rigid to our deformable registration (Table 2, green) and conversely a perceived degradation when transitioning from our deformable to rigid registration (Table 2, red). When the method was held constant through the registration transition, we see almost no perceived improvement (Table 2, yellow). Performing a pair-wise t-test over the 20-patient cohort assuming correlated samples and in which a pair consisted of comparing the average patient’s rigid-to-deformable display evaluation score to that of the deformable-to-rigid display evaluation score, there was found to be a strong statistically significant difference (p < 0.001, with a difference in the mean scores of 2.95). Figure 1 is an example case from the study. The top panel shows a guidance display, a model view, and a surface fit when rigid registration is used, respectively. The bottom panel shows the deformable counterpart. Misregistration as much as +/- 15 mm is shown in top panel with marked decrease in surface fit error in lower panel. Figure 1 left-column-top guidance display shows misaligned crosshair deep within liver (stylus was physically touching liver) while in the corresponding lower corrected panel, we see the crosshair nicely back on the surface. CONCLUSIONS: The above bystander study is unique among the guidance community and suggest that it represents a novel testing framework for assessing image-to-physical registrations in real time. In addition, our sparse-data driven deformable registration provides significant perceived improvement on localization. ACKNOWLEDGEMENT: This work is supported in part by NIH-R01CA162477 and NIH-T32EB021937.

ABSTRACT. In 3 patient cases, intraoperative stereovision (iSV) image pairs were acquired after dural opening and at the end of resection to generate surface displacement maps. These data were assimilated into a biomechanical model to create updated MRI (uMR) that correspond to the surgical scene, and intraoperative MRI (iMR) was acquired and compared to uMR. In addition, since iMR is typically acquired at the end of resection and multiple iMRs are not available due to patient safety concerns, we developed a large animal glioma model where iMRs can be acquired at multiple surgical stages for uMR validation. We have also adapted image updating to minimally-invasive settings such as tumor ablation and deep brain stimulation (DBS). We generated updated CT (uCT) in 16 deep brain stimulation (DBS) cases and quantified TREs of deep brain structures by comparing to post-operative CT (post CT). Surface-to-surface distance (SSD) in the surgical cavity between uMR and iMR was 2.4±1.8 mm, 1.6±1.0 mm, and 2.6±1.6 mm, respectively, which was significantly better than preoperative MRI (pMR) where SSDs at end of resection were 7.7±4.6 mm, 2.9±2.7 mm, and 8.4±3.9 mm, respectively. We developed and optimized pre-implantation immune-suppression therapy sufficiently to grow solid tumors from implanted U-87 MG cells in the brains of swine, and show gross brain and corresponding MRI slices from 2 example animals. In the 16 DBS cases, TREs for uCT were 1.3±0.3 mm and 1.8±0.8 mm at subsurface anterior and posterior commissure points, respectively, and image updates were possible for 3 cortical shift patterns: asymmetrical, symmetrical, and minimal. In sum, our image updating system to produce uMRs is functional in open-cranial and minimally invasive procedures. Initial steps to validate our results by comparing end-of-resection uMRs with iMRs, developing a glioma model for validation at multiple surgical stages, and adapting the image updating system to DBS cases are encouraging.

ABSTRACT. There is widespread consensus that pediatric exposure to ionizing radiation for medical purposes should be minimized whenever possible. The “Image Gently” campaign from The Alliance for Radiation Safety in Pediatric Imaging was launched in 2008 and has gained widespread momentum in pediatric hospitals across the US. Therefore we are developing methods to facilitate diagnostic and therapeutic procedures under MRI and ultrasound guidance to provide radiation free options for our pediatric patients. In this abstract we briefly describe some of these efforts.

ABSTRACT. Purpose. Oral and Maxillofacial (OMF) surgery is a discipline that manages surgical correction of jaw discrepancy and malignant diseases of the head and neck. OMF residents have little to no training with osteotomy saws before direct patient contact. To bridge this gap, we have developed a high-fidelity virtual reality (VR) osteotomy simulator to aid the acquisition of the psychomotor and visual perception skills necessary to perform Sagittal Split Osteotomy (SSO) surgery. Methods. We have developed a fully virtual training environment that renders the anatomy, tools, haptics, collision detection and physical responses existing in the SSO procedure. The environment includes a VR Head Mounted Device (HMD), a physics simulation engine that creates high fidelity haptic feedback, a saw/bone interaction algorithm that can render the experience of performing mandibular osteotomies and a framework that collects performance-based metrics (PBMs) for measuring the precision of the surgical cuts and skills progression. We have built this infrastructure using iMSTK[1] as physics and rendering engine. Our virtual training scenario includes a highly accurate model of the different craniofacial structures of a patient that underwent SSO surgery obtained from Cone Beam Computed Tomography (CBCT). Each point in the mandible mesh embedded in the model of the bone in the model is initialized with a bone density value that is proportional to the intensity values of the CBCT scan. Resulting collisions from the interaction of reciprocating saw and mandible are detected at every frame using a custom localized collision detection. A novel bone-saw interaction algorithm computes the topology changes as well as the forces involved in the bone cuts for the osteotomy [2]. Our system renders the collision force-feedback through a customized haptic device that mimics the physical reciprocating saw. Finally, we record three PBMs that measure that the cuts follow a ground truth path and penetrate a single cortical layer without colliding with the alveolar nerve bundle. Results. In order to ensure we have produced a realistic osteotomy experience we asked residents and experienced OMF surgeons to rate our system. Trainees were asked to perform a SSO osteotomy in our system and to reply some questions that investigated (1) ease of operating the system; (2) realistic appearance of the scenario; (3) adequacy of force feedback; (4) mental and physical demand required when using the interface; and (5) potential areas of improvement. Our results include an average of 3 out of 5 in the Likert scale indicating our system renders a close-to-real osteotomy experience. Conclusions The mandibular osteotomy is the most critical step in the SSO procedure. Surgical errors can affect outcomes in SSO surgery, including excess bleeding at surgery, post-surgery nerve dysfunction, and surgical failure due to incorrect bone cuts. Our initial experiments indicate that the simulation system developed shows great promise towards training, acquiring the surgical skills necessary to perform SSO accurately without additional risk for patients. Future work contemplates improving simulation fidelity and speed, graphical user interface, hardware interface and training data performance handling of our current system, as well as performing a full set of learning-curve experiments via the collected PBMs. Funding Source. This work was supported by the National Institute of Health (NIH) National Institute for Dental and Craniofacial Research (NIDCR) grant R43DE027595 High fidelity virtual reality trainer for orthognathic surgery.

ABSTRACT. Purpose: Minimally invasive interstitial therapeutic ultrasound (TU) enables conformal control of the boundary of a thermal ablation region in the brain. However, precise placement and real-time monitoring of thermal dose are critical to ensuring effective therapy. In this work, a seven degree-of-freedom robotic system is utilized for in-bore MRI-guided robotically assisted (MRgRA) placement of an ACOUSTx conformal TU applicator from Acoustic MedSystems. TU ablation is performed under MR-based thermal imaging (MRTI) generating a real-time temperature map and delivering live thermal dose calculations within the target region. This enables the geometry of the ablated region to be supervised during intervention and the procedure stopped when the desired ablation zone size is reached. In our previous work, we demonstrated the capabilities of the NeuroRobot system and workflow. The purpose of our presented work is to demonstrate and evaluate the MRgRA conformal TU brain tumor ablation platform and thermal monitoring system during pre-clinical survival porcine trials.

Methods: The system is portable and setup prior to the procedure. This entails configuring the robot and its base platform on the MRI scanner bed, placing the controller beside the scanner in the room, connecting the robotic platform to a fully shielded custom control box via a 15-foot cable, and connecting the controller to a monitoring PC in the console room via fiber optic network. The custom MRI robot controller comprises up to ten custom-tailored daughter cards for control of various types of piezo-electric actuators used by the NeuroRobot. The monitoring PC interacts with the controller through a web-based operators console. The controller communicates via OpenIGTLink over the fiberoptic network to Slicer and the TheraVision software for planning, delivering TU, and monitoring the procedure. During the system setup, registration is performed via a fiducial mounted on the base platform. Then an entry point and an intra-cranial target point are selected. The robot aligns the probe along the planned trajectory, a cannula is inserted through the burr hole along the robotic guide, and the TU probe is attached to the robot and inserted to the specified depth. MR images confirm placement, the MRTI application configures the scanner to acquire a stack of cross-sectional images normal to the probe axis, and ablation is controlled via the TheraVision ultrasound ablation system. A real-time thermal dose map monitors the ablation region.

Results: Lab-based experiments using motion capture demonstrated a free-space targeting accuracy of 1.37mm. Phantom studies in MRI exhibited a targeting accuracy of 2.37mm. Minimal degradation of image quality with a mean SNR loss of 2.9% was observed with the robot powered on within the MRI (the typical scenario), and a loss of up to 10.3% with the robot moving during imaging, and image distortion was <0.2mm. Six porcine trials have been conducted with the fully integrated system for MRgRA conformal brain tumor ablation under MRTI, four acute and two survival. Thermal monitoring results were compared to gross examination. The survival swine showed no neurological sequelae at two weeks post procedure, and T2 and FLAIR images were accurate reflections of the ablation volume 1-week post-procedure.

Conclusions: We present an MRI-guided robotic system for conformal brain tumor ablation with real-time thermal dose monitoring. Preclinical survival studies have shown promise with intra-operative and post-operative imaging correlating with histological examination.

Funding Source: Supported by NIH NCI Academia-Industry Partnership R01 CA166379

ABSTRACT. Purpose: Neonatal hypoxia-ischemia continues to be a significant problem, despite therapeutic advances of hypothermia. We have previously shown that intra-ventricular delivery of human glial progenitors (GPs) at the neonatal stage is capable of replacing host abnormal glia cells, rescuing the life span of dysmyelinated mice. However, the small size of the murine brain does not allow proper investigation of the transplantation challenges related to high-volume ventricles and long transport distances for stem cells as encountered in the human brain. Our long-term goal is to study the potential benefits of complete glia replacement in a large animal (piglet) model of neonatal hypoxia-ischemia. Because the cerebral ventricles would be an attractive gateway to introduce cells to vast brain areas across the entire neuroaxis, we used bioluminescence (BLI) and real-time MRI to investigate the potential variables that could optimize the biodistribution of injected glial progenitors within the ventricular system while minimizing their outflow to the intrathecal space.

Methods: 3D model preparation: A 3D model of the piglet cerebral ventricles was rendered based on T2-weighted MR images (11.7T Bruker Biospin) and post-mortem brain dissection and subsequently 3D printed (Stratasys). Stem cell infusion: Luc(+)GPs were used for all experiments, and in some cases there were labeled with SPIO (Molday ION Rhodamine B, Biopal). They were infused into our 3D printed phantom of cerebral ventricles using a port mimicking a trajectory typically used for shunt placement, which is a frequent pediatric neurosurgical emergency. BLI: The bioluminescent GP signal was measured using an IVIS Spectrum/CT instrument (Perkin Elmer). BLI was quantified by drawing of regions of interest (ROIs), with data expressed as photon flux (p/sec). MRI: An undersampled radial FLASH pulse sequence was used with a 3T Prisma Fit MR scanner (Siemens). Statistical Analysis: Result were calculated and subjected to statistical analysis using multivariate regression (PROC MIXED, SAS 9.4).

Results: BLI revealed that lower injection speeds and lower volumes minimize cell outflow outside the ventricular system, while having little effect on the distribution within the ventricular system. Real-time MRI demonstrated that SPIO-labeling significantly alters rheological properties of glial progenitor suspension such that, even at high speeds and high volumes, outflow beyond the ventricular system was reduced. Further detail analysis revealed that the cell origin (human vs. mouse), type of catheter (one-hole vs two-sided hole) and length of the catheter had less effect on the distribution of GPs.

Conclusions: Infusion speed, volume and SPIO labelling strongly influence the biodistribution of glial progenitors injected in a 3D model of piglet cerebral ventricles. Use of 3D model was essential for cost-effective optimization of the intraventricular delivery before it is pursued in large animal studies and clinically translated.

Funding Sources- This study was supported by R01 NS091100

ABSTRACT. Focused ultrasound combined with microbubble-induced blood-brain barrier disruption (FUS-BBBD) is a promising technique for noninvasive and targeted brain drug delivery with clinical trials currently on-going. This study introduced the use of passive cavitation imaging (PCI) for predicting the location and concentration of nanoclusters delivered by FUS-BBBD.

ABSTRACT. To make use of preoperative segmentation results for intraoperative navigation, registration between pre- and intraoperative data is necessary. However, surgeries are often guided by micro- or laparoscopic video, which can only provide two-dimensional (2D) videos. Direct registration between 3D and 2D data is difficult and inaccurate. The goal of this work is to develop a method to reconstruct 3D tissue surface from the stereo micro- or laparoscopic video in real-time, which can greatly facilitate registration between intraoperative 2D video and preoperative MR or CT imaging.

ABSTRACT. Nerve damage plagues surgical outcomes, significantly affecting post-surgical quality of life. Surprisingly, no method exists to enhance direct nerve visualization in the operating room, and nerve detection is completed through a combination of palpation and visualization when possible. Fluorescence image-guided surgery offers a potential means of enhanced intraoperative nerve identification and preservation. To date, a variety of nerve specific fluorophores have been tested in preclinical models, however a clinically approved near infrared (NIR) nerve-specific contrast agent does not yet exist. Herein we report our efforts to synthesized a focused library of systematically-modified far red to NIR fluorophores to define the factors that modulate a fluorophore’s nerve specificity. A library of 66 novel compounds was synthesized and characterized for optical and physicochemical properties. 41 of the novel compounds had maximum emission wavelengths that fell within the NIR imaging window (650-900 nm). The nerve-sparing radical prostatectomy is a surgical procedure that could benefit from fluorescence image-guided nerve identification. Although the nerve-sparing surgical technique was developed over 30 years ago, nerve damage following radical prostatectomy is reported in up to 60% of patients post-surgery. To facilitate clinical translation of fluorescence image guided surgery to the nerve sparing prostatectomy, a direct/topical administration methodology has been developed that allows selective nerve highlighting with a significantly lower fluorophore dose than systemic administration, where large animal studies have confirmed the technique’s translatability. Notably, when scaled by body surface area the optimal dose for direct administration falls within the microdosing range, permitting first-in-human clinical studies under an exploratory investigational new drug (eIND) application, making these studies more financially achievable. Novel gel-based formulation strategies have been investigated and optimized to enable more selective staining through decreased background accumulation and enhanced tissue penetration for identification of buried nerves. Additionally, clinically relevant formulations have been developed for systemic administration using intravenous injection for applications in which direct administration would not be ideal. Given the interplay between novel NIR nerve-specific contrast agent development and clinically relevant formulation strategies for direct and systemic administration, translation to the clinical should be feasible within the next five years.

ABSTRACT. Prostate cancer is known for its high survival rate when it is localized due to slow tumor growth, but the survival rate drops according to the initiation of metastasis and accelerated tumor growth. Although the image-guided needle biopsy has been the gold standard for prostate cancer screening as well as active surveillance, the procedure is non-trivial and fraught with complications including pain, bleeding and infection. Thus, there is a strong need of developing a noninvasive approach to detect prostate cancer and to repetitively monitor its growth. We propose a molecular photoacoustic (PA) imaging system that depicts targeted prostate-specific membrane antigen (PSMA). The PA agent contrast showed an enhancement on the PSMA+ tumor comparing before injection and after 24-hour, and the PA agent contrast difference was observed by comparing PSMA+ and PSMA- tumors at 24-hour post-injection. These results demonstrate the potential of PA imaging to characterize prostate tumors for early detection and active surveillance.

ABSTRACT. Vibroacoustography (VA) is a promising non-invasive imaging modality that contrasts tissue regions by identifying disparities in viscoelasticity. Our research explores the application of this technology to head and neck cancer. Looking forward, an intra-operative instrument to quantitatively image the mechanical properties of oral cancer tissue may increase precision in determining true resection margins.

ABSTRACT. CBTF is a new x-ray imaging modality designed for image-guided interventions. This paper is an overview of its underlying technologies: a novel imaging system and advanced reconstruction and visualization techniques. We include a review of existing results, and discuss the pre-clinical validation methods required to assess the viability of CBTF for different clinical applications. While in its early developmental stages and with performance still improving, current results are encouraging for the use of the modality in orthopaedic applications. Pre-clinical and clinical research is still required to conclude on the benefits and drawbacks of CBTF in orthopaedics. Other clinical applications that could benefit from this modality such as biopsies, tumor resections, pain management, or cement augmentations are yet to be explored.

ABSTRACT. Due to its wide availability, real-time capability and low cost compared to other modalities, ultrasound (US) imaging plays a central role in IGT. Knowing the 3D position and orientation of the imaging probe allows acquired 2D slices to be properly oriented in space, for example to aggregate 2D slices into 3D volumes and/or facilitate image fusion. The two most common ways of tracking US probes are optical and electromagnetic (EM) tracking, but both have serious limitations as they are arguably burdensome to use, not well-suited to the MRI environment, and relatively expensive. We propose the use of US-based sensors attached to the skin to ‘spy’ on US scanners, and converting spied signals into tracking data using a convolutional neural network (CNN).

ABSTRACT. Purpose: Endovascular catheter-based procedures under MRI guidance for neurointerventional applications are inherently difficult. One major challenge is tracking the tip of the catheter under MRI, as standard fabrication methods for building semi-active markers are rigid and bulky. We used aerosol jet deposition to print a complete LC circuit using the geometry of a double helix inductor¹on a polymer catheter for interventional MRI use at 3T.

Methods: A catheter was constructed using a custom 6Fr catheter PTFE substrate, and polyethylene ether ketone (PEEK) braiding (Penumbra Inc., Alameda, CA). The conductive traces of the double helix¹were printed with a 250 μm trace width, 10 turns with a 22.5° pitch, and a 5–7 μm trace thickness. The capacitor plates were printed measuring 180 mm long, with 280–300 μm trace width, with a similar 5–7 μm trace thickness, and 200 μm separation of the plates (Fig. 1a). All conductive prints were made using a water-based silver flake inkand polyimide dielectric ink (Aerosol Jet^®300P system, Optomec, Albuquerque, NM)  (Fig. 1b). The catheter was placed in a water phantom and was oriented parallel or perpendicular to B₀at 3T (Discovery MR 750w, GE Healthcare) using an 8-channel cardiac coil. A balanced steady state free precession (bSSFP) sequence (TE/TR = 1.7/4.63ms, 30 mm²FOV, slice thickness 20 mm, matrix 384 × 384) was acquired with flip angles 5°, 15°, and 75°. A 5° Bloch-Siegert B₁+ map^2,3was also acquired (TE/TR = 13.4/28ms, 30 mm²FOV, slice thickness 20 mm, matrix 128 × 128). The mean signal of a manually drawn ROI of the marker was measured and compared with the nearby water signal to measure the relative contrast ratio RCR = (C_marker– C_water)/C_water.                                                                

Results: The low flip angle bSSFP sequence (5° and 15°) shows the signal amplification of the markers relative to the background signal, in comparison with the high flip angle (75°) sequence (Fig. 1c). In the parallel orientation, the RCR measured 5.26, 4.07, and 2.38 for 5°, 15°, and 75° flip angles respectively, and 0.53 for the 5° B₁+ map (Fig. 1c). In the perpendicular orientation, the RCR measured 1.6, 1.01, and -0.02 for 5°, 15°, and 75° flip angles respectively, and 0.64 for the 5° B₁+ map (Fig. 1d).

Conclusions: This preliminary data suggests that fabrication of complete 3D printed LC circuits for use as tracking markers on catheters is possible and that these markers can exhibit good tracking characteristics at 3T.

References: 1. Thorne, B. R. et al.ISMRM.21,2327 (2014). 2. Khalighi, M. M., et al. MRM68,857–862 (2012). 3. Sacolick, L. I., et al. MRM63,1315–1322 (2010).

ABSTRACT. This work demonstrates non-contrast-enhanced MRI for the purpose of imaging necrotic RF ablations. The conventional EP suite lacks the tools to verify permanent conduction block after ablation. Transient injury may show temporary relief from arrhythmia and patients are often discharged with lesion gaps which are associated with recurrence. Pre-clinical results show robust and selective enhancement of necrotic lesions, corroborated by gross pathology. The proposed imaging technique may be beneficial to verify completeness of ablative treatment.

ABSTRACT. Purpose: Accurate tracking of interventional device position inside the MRI scanner is of paramount significance to ensure surgical safety. There are two methods to track device positions directly within the MR environment. A passive approach relies on identifying image susceptibility artifact of the device. For this, a volumetric MR scan has to be performed, which is a time-consuming process and the performance relies on the accuracy of artifact segmentation method. An active approach relies on radio-frequency coils to achieve the 3D localization at faster rate and higher resolution [1, 2]. We hypothesize that the accuracy of robot assisted interventions could be improved via the use of MRI active tracking coils.

Figure 1. A: The 1-DOF MR-conditional robot for needle insertion procedures [3]; B: schematic diagram of the MRI active tracking coil [2]; C: MR-conditional robot inside a 3T scanner; D: coil mounted on a plastic needle with the same size of breast core biopsy needle; E: Insertion distance with optical encoder feedback and active tracking coil feedback. Methods: A 1-degree of freedom robot (figure 1) was constructed to validate the proposed method. The robot was designed to deploy a core breast biopsy needle under MRI guidance inside a 3 Tesla Achieva Philips Whole body MRI scanner. A cutout was made at the needle tip to hold the MRI tracking coil. The tracking coil spatial information was computed in the MR host using rapid interleaved 3D linear navigator segments and stored in a text file. A custom program developed in C# was used to read the text file in real time and establish the communication between the host and robot control PC. The real-time position was sent to the xPC target via USB-serial communication (Baud rate: 19200) for robot close loop control. Results: The insertion experiment was performed in a gel phantom. The accuracy performance with active tracking technique was compared to the one with joint space optical encoding method. The desired insertion length was chosen as 38mm, while the actual insertion with active tracking coil and joint space optical encoder feedback was 37.99mm and 38.82mm, respectively. Conclusions: The proposed active tracking coil based close loop control could potentially improve the needle insertion accuracy in the MRI-guided robotic interventions. References: 1. Y. Chen, TMech, 2016; 2. S. Sengupta, MRM, 2014; 3. Y. Chen, TMech, 2017;

ABSTRACT. Purpose There exists a pressing medical need for non-invasive optical tools to help surgeons intraoperatively identify tissue and determine precise tumor margins. Our team presents a novel technology, termed dynamic optical contrast imaging (DOCI), in a mobile system designed for facile use in the operating room. The objective of this study is to demonstrate the utility of DOCI in reliably and accurately identifying neoplastic tissue in head and neck surgery.

Methods A prospective series of patients diagnosed with head and neck squamous cell carcinoma (HNSCC) at the David Geffen School of Medicine at UCLA was examined. Oral and HNSCC tumors and their surrounding tissues in the surgical bed were collected and fluorescence decay images were acquired using a wide-field DOCI system. Samples (46 patients) were subsequently processed for standard histological assessment. Contrast is created by taking measurements during the decay process where successive lifetime measurements occur in sync with the expected decay times. These measurements are then subtracted and normalized to each other to create a time map of relative fluorescent decay. Multispectral relative fluorescence decay values were quantified for both in vivo and ex vivo settings.

Results Our system extracts relative fluorescence decay information in a surgically relevant field of view with a clinically accessible acquisition time < 30 seconds. Ongoing human in-vivo and ex-vivo trials suggest DOCI is a well-tolerated, low-cost, and sensitive tool for differentiating diseased from normal tissues throughout the head and neck, in addition to the oral cavity. Furthermore, in-vivo intraoperative results demonstrate significant potential for image-guided detection and resection of oral cavity squamous cell carcinoma (OSCC).

Conclusions This study demonstrates a promising new imaging modality capable of generating intrinsic contrast between tissue types without using dye or ionizing radiation. The continued development of this system may aid in patient screening, reduce unnecessary biopsies, delineate margins for tumor resection, provide guidance in choice of biopsy sites, and preserve healthy tissue to increase the postoperative functionality and quality of life of the patient.

Funding Source(s) UCLA Medical Scientist Training Program (grant T32GM008042) National Cancer Institute - R01

ABSTRACT. Purpose: Intra-operative identification of tumors remains challenging for surgeons when performing transoral robotic surgery (TORS) for oropharyngeal carcinoma. Currently, surgeons rely heavily on visual cues and preoperative images to develop surgical plans for resection. Current methods of preoperative marking are limited, as blood and tissue may obscure the view and can fade with time. This makes them unsuitable for long-term treatment protocols which may use chemotherapy or radiation prior to surgery. We are aiming to develop and evaluate a clearly visible, long-lasting, precise, intra-operative tumor marking method for guiding TORS. Methods: Each marker is comprised of a solution of clinically approved Indocyanine green (ICG), biocompatible cyanoacrylate and acetone. A light source with a peak emission of 760 nm excites the near infrared fluorescent (NIRF) marker, which fluoresces at a peak wavelength of 806±5 nm. By applying a band-pass filter (845±27.5 nm) in front of the NIR camera, we can specifically observe the excited fluorescent light from the NIRF markers. The surgeon manually marks the location of the tumor with 0.03 ml subcutaneous injections of NIRF solution with a 26G bevel needle. When the NIRF solution contacts tissue, it quickly hardens into a bead about 6 mm in diameter and the NIR signal penetrates the tissue. We designed and built a camera stand that protects the samples from external light to achieve constant lighting and imaging conditions (Fig 1a). 3 cadaver porcine tongues (Fig 1b) and 2 cheeks are used for this preliminary study with 4 NIRF markers implanted underneath the skin surface for each sample. Pictures with the NIR camera were taken for 26 days from directly above the samples (Fig 1c). The signal intensities and areas of each marker were measured on collected images (Fig 1d). Results: In all experiments, the intensity threshold for segmentation of the NIR markers was set to 75 grayscale units (0-255). The mean signal pixel intensity over 26 days of study dropped from 116±17 to 108±15 for tongues (N=12) and from 135±25 to 120±18 for cheeks (N=8) with the majority of decrease happening over the first 2 days. The average background intensity equaled 21 and 16 for tongues and cheeks respectively, resulting in a signal to noise ratio (SNR) greater than 5 for the entire period of testing. The marker area reductions follow a pattern similar to the intensity, decreasing from 38±13 to 31±13 and from 48±19 to 26±8 mm2 for tongues and cheeks respectively. However, the decreases in both intensity and area were not found to be statistically significant based on the ANOVA test (p=0.28 and p=0.65 while alpha value equals 0.05, respectively). What’s more, the camera’s resolution exceeds millimeter accuracy at 0.12 mm/pixel, and the frame rate is 19Hz, which satisfies the real-time tracking requirements for TORS in future work. Conclusions: The NIRF markers and imaging system in this study enable identification and tracking for a period of at least 26 days in cadaver porcine tongues and cheeks. Considering the high frame rate of the camera, penetration of NIR light, and high SNR, this marking strategy has the potential for real-time intraoperative tracking of tumors for long-term treatment protocols. Funding Source(s): This work is supported by the National Institute of Biomedical Imaging and Bioengineering of the NIH under award numbers 1R01EB020610 and R21EB024707.

ABSTRACT. Automated Image-Guided Pedicle Screw Placement Using a Commercial Robot Alexander D. Smith, Jacob Chapin, Joao Lopes, Andrew F. Hall Department of Biomedical Engineering, Parks College of Engineering, Aviation, and Technology Saint Louis University, Saint Louis, MO Purpose The purpose of this study was to evaluate the initial performance of an image-guided robotic surgery system, developed using a self-localizing commercial robot, to automatically place screws in bone tissue. Current state-of-the-art surgery systems use robots only to locate and align the tools; the procedure being left to the surgeon. Furthermore these systems require integrated device localization systems. We sought to automate the entire process of screw placement, using the localization capability inherent to the robot. Methods The system was based on a KUKA LBR iiwa R800 industrial robot (Kuka Robotics, Augsberg, Germany), the 3D Slicer open-source medical image visualization platform (www.slicer.org), and a Schunk PRH Rotary Unit (Schunk GmbH, Lauffen/Neckar, Germany). Robot operation was programmed in Java via the robot’s integrated development environment. The rotary unit was controlled by the robot software via a PROFIBUS interface. Fiducial marker and screw planning data were transferred manually from 3D Slicer to the robot. The robot end effectors (awl, drill bit, screw holder driver) were built from a combination of commercial spine tools, industrial hardware and custom 3D-printed parts. Image/robot registration was accomplished by hand-guiding the robot to the fiducial markers, where they were localized. First, registration and navigation were tested in a phantom. The end effector was registered to the phantom and then navigated through a bone entry point to a destination point in four different locations. Next, three previously frozen bovine femurs were instrumented with four fiducial markers each. A C-Arm CT image (Siemens, Erlangen, Germany) was then captured for each bone. The images were imported into 3DSlicer, where the fiducial markers were identified and locations (insertion and destination) for two screws were defined. Two screws were placed in each bone as follows: The robot was hand-guided to each of the four fiducial markers (Fig. 1-a). The robot software used this data to register the image data to the robot coordinate system. For each screw, the robot used an awl followed by a drill bit to create the channel for the screw, and then picked up and placed the screw automatically using a magnetic driver (Fig. 1b). Tools were exchanged manually. After the screws were placed, another C-Arm CT image was taken and imported into 3D Slicer. The insertion and destination locations of the screws were identified. The post-procedure images were then registered to the pre-operative images. The distance between each planned and actual insertion and destination point was then computed. Results Registration/navigation testing was completed twice (n=8). The destination error was 0.55±0.30 mm. For the six screws placed, the mean insertion depth was 42.3±1.9 mm, the insertion depth error was 1.86±1.41 mm, the insertion location error was 1.50±0.63 mm and the destination location error was 4.45±1.05 mm. The screw placement measurements are approximately a factor of 2 larger than those reported for clinical robotic assist devices, and for other experimental robotic screw insertion systems. Several areas of improvement have been identified for this system and experimental set up. The fiducial markers were steel, and had significant image artifacts, adding to registration error. The pilot hole made by the awl was not ideal for automatic engagement of the drill, and the 3D printed drill tool was not completely rigid. Conclusions This study demonstrated the feasibility of performing a critical component of intervertebral spinal fusion using a self-localizing industrial robot, without the requirement for a device localization system. Further development in both the tools and technique should improve the accuracy. Funding Source There were no funding sources.

ABSTRACT. Purpose: A minimally invasive surgical technique, video assisted thoracoscopic surgery or VATS, has emerged over the last 15–20 years as the de facto standard of care for lung tumor resection, comprising ~60% of surgical cases in the US [1]. Surgeons insert tools through small ports between the ribs, one tool being a thoracoscope used to view the inside of the chest.

The surgical challenge in VATS is to localize lung tumors in the thoracoscope view, in order to resect the tumors with safe margins while sparing healthy tissue. Here we propose the use of augmented reality (AR) for image guidance, with a focus on the particular challenges of lung surgery and our corresponding solutions. Applied to lung surgery, existing AR approaches [2] face key limitations in tissue tracking: (1) irregular motion from surgical manipulation (stretching, grasping), (2) physiological motion, (3) tool occlusion (4) high degree of lung tissue deformation, and (5) discerning lung tissue from the surrounding anatomy (e.g., pericardial sac, ribcage).

Methods: Our augmented VATS solution (Fig. 1a) overlays graphical objects, such as a tumor (Fig. 1c), on the thoracoscopic video. After an initial overlay registration, background subtraction [3] and multiple HSV masks are used to identify and remove the background and tools (Fig. 1b). These masked frames supply a dense optical flow algorithm [4] with flow smoothing. Tracked motion is then fit to an affine model using least-squares minimization, and the overlay graphic is transformed accordingly. Overlays are updated for each new video frame to follow the motion of its underlying tissue base.

Results: The pipeline was tested with video of size 720×486 px at 24 fps. A computational rate of 10 Hz was achieved using a C++ implementation on a GPU. In our in vitro setup (Fig. 1a), the overlay was qualitatively assessed to adhere to anatomical features of a porcine lung specimen under simulated surgical motions such as gross manipulation, grasping, and stretching.

Conclusions: Our experimental setup and methods demonstrate potential in recreating and addressing the particular challenges of AR for lung surgery. Future work includes quantifying the performance, integrating preoperative imaging, and developing clinically viable workflows.

References: [1] Healthcare Cost and Utilization Project (HCUP), https://www.hcup-us.ahrq.gov. [2] H. Elhawary, Int. J. Med. Robot. 2011. [3] A. Godbehere, Am. Control Conf. 2012. [4] Brox, Eur. Conf. Comput. Vis. 2004.

Funding Source: This work was internally funded.

ABSTRACT. Purpose: Our long-term goal is to develop a microwave ablation (MWA) system for precise targeting of lung tumors, integrated with a clinically established approach for image-guided bronchoscopic transparenchymal nodule access and treatment planning system. The rationale for this research is that image-guided bronchoscopic MWA, combined with comprehensive planning and monitoring, may enable safer and more effective therapy than current percutaneous ablation techniques. Methods: We employed computational models to design a 2.45 GHz catheter-cooled MWA antenna (1.5 m length, 1.8 mm O.D.), suitable for introduction to lung tumors via the working channel of a flexible interventional bronchoscope. Proof-of-concept applicators were fabricated and experimentally characterized in ex vivo liver tissue – an established model for characterizing thermal ablation devices – with 30 – 45 W applied for 5 – 10 min (n = 4 at each combination). A prototype treatment planning system was developed to find the optimal path to deliver the applicator to the tumor and predict the ablation zone based on input generator parameters. Results: The width and axial ratio (a measure of the sphericity of thermal ablation zone) of the observed ablation zones ranged between 19 – 32 mm and 0.73 – 0.87, respectively, similar to those that can be achieved with needle-based microwave ablation applicators currently in clinical use for percutaneous ablation of lung tumors. Figure 1 shows an example optimal path along with the point of entry and the ablation zone suggested by the treatment planning system. Conclusion: We have developed a proof-of-concept 2.45 GHz flexible MWA applicator suitable for treatment of lung tumors via a bronchoscopic approach, and a treatment planning platform integrated with the Archimedes guidance and navigation platform. Efforts are underway for detailed characterization of the ablation system in animal lung in vivo. The proposed system may enable precise delivery of thermal ablation for treatment of early-stage lung tumors. Funding Source: NIH/NCI grant, award number R01 CA218357.

ABSTRACT. Purpose. To create an efficient, accurate, and radiation-free stereovision imaging device that facilitates image registration and surgical navigation during open spine surgery. The stereovision device is designed to acquire intra-operative profiles of the exposed spine, which can be used to register and update pre-operative CT of the spine acquired in a supine position to compensate for vertebral posture and alignment changes when the patient is in prone during surgery.

Methods. The hand-held stereovision (HHS) device (Figure 1) includes two high-definition cameras (C920, Logitech, Lausanne, Switzerland) that acquire image pairs of the surgical field (1920 x 1080 pixels, 30 frames per second) from which a three-dimensional (3D) surface is reconstruct-ed. We calibrated the stereo parameters by taking images of a checkerboard (Figure 2) from different locations and angles. An active tracker (Medtronic, Louisville, Colorado) was rigidly attached to the HHS and spatially calibrated using a tracked stylus to obtain the transformation between the device’s field of view and the tracker (Figure 2).

Accuracy of the HHS was evaluated in an ex-vivo swine specimen in the operating room (OR). Leibinger mini-screws were implanted in the tips of each spinous and transverse process from L1 to L6 (N=18). Their ground truth locations in the physical space (Figure 3 red points) were acquired with a tracked stylus (StealthStation S7, Medtronic, Louisville, Colorado). Stereo images of the exposed spine were acquired to reconstruct the exposed vertebrae (Figure 3). The reconstructed screw locations were localized and compared to their tracked counterparts to evaluate accuracy. Data was acquired using Labview (National Instruments, Austin, TX) and processed in Matlab (MathWorks, Natick, MA).

Results. The mean calibration reprojection error was 0.33 pixel. The spatial tracking error was 0.56 ± 0.25mm. A 3D profile was reconstructed within 3s for each image pair with a point-to-point accuracy of 2.06 ± 1.21mm across 3 surfaces and 15 screws.

Conclusions. These preliminary results are promising and suggest that our HHS device has the potential to provide data from the intraoperative surgical field to enable image guidance. Our next step is to evaluate potential of the stereovision data to compensate for spine intervertebral motion resulting from patient posture changes between pre-operative (supine) imaging and intra-operative (prone) surgical positions.

Funding Source: R01 EB025747-01 and R21EB020859-01A1 awards from the National Institute of Biomedical Imaging and Bioengineering (NIBIB).

ABSTRACT. Authors: William W. Phillips, MD,a Christopher S. Digesu, MD,b Kathleen D.Weiss, MD,a Krista J. Hachey, MD,c Yolonda L. Colson, MD, PhDa

Purpose: To report the first long-term outcomes using near-infrared (NIR) image-guided sentinel lymph node (SLN) mapping in non-small cell lung cancer (NSCLC).

Methods: Retrospective analysis of NSCLC patients enrolled in two prospective phase 1 NIR-guided SLN mapping trials, including an indocyanine green (ICG) dose-escalation trial. All patients underwent NIR imaging for SLN identification followed by multi-station mediastinal lymph node sampling and pathologic assessment. Disease-free (DFS) and overall survival (OS) were compared between NIR+ SLN patients (SLN identified) and those without (Non-SLN group). Survival probabilities utilized Kaplan-Meier analysis.

Results: Forty-two patients undergoing peritumoral ICG injection were assessed. Median follow-up was 44.5 months. SLNs were not identified in 19 patients prior to ICG dose optimization or due to imaging failures and constituted the Non-SLN group. The subsequent SLN group included 23 patients. SLN pathology was concordant with overall nodal status in all patients. Sixteen SLN patients were deemed pN0 and no recurrences were identified; whereas, 4 of 15 pN0 Non-SLN patients developed nodal or distant recurrent disease. Comparing SLN vs. Non-SLN pN0 patients, the probability of 5-year OS is 100% vs. 70.0% (log-rank P=0.062) and 5-year DFS (Figure 1) is 100% vs. 57.9% (log-rank P=0.0009), respectively. Among 11 pN+ patients, 7 were in the SLN group with 3 patient’s disease confined to the SLN alone.

Conclusions: Patients with pN0 SLNs exhibited favorable disease-free and overall survival. This preliminary review of NIR SLN mapping in NSCLC suggests that pN0 SLNs may better represent true N0 status. A larger clinical trial is planned to validate these findings.

Funding Sources: This work was supported by National Institutes of Health R01CA131044 and the Brigham and Women’s Hospital Advanced Training in Surgical Oncology T32 Fellowship (5T32CA009535) for C.S.D. and K.J.H.

From the aDivision of Thoracic Surgery, Brigham and Women’s Hospital, Boston, Mass; bDepartment of Surgery, Beth Israel Deaconess Medical Center, Boston, Mass; and cDepartment of Surgery, Boston Medical Center, Boston, Mass. Drs W.W.P. and C.S.D. contributed equally to this work.

ABSTRACT. Purpose: Cone-beam CT (CBCT) and diffuse optical tomography (DOT) are emerging 3D imaging modalities for surgical guidance. Here, we develop a hybrid CBCT-DOT system using surgical navigation technology and computational light transport models.

Methods: An optical tracking system provides real-time localization of laser and camera devices used for non-contact DOT acquisition. CBCT imaging is used to generate tetrahedral meshes for a finite element method implementation of DOT (NIRFAST). Fluorescence camera images are mapped onto the mesh surface using a computational model of free-space light propagation. Structural data from CBCT is incorporated directly into the optical reconstruction process using Laplacian-type regularization (“soft spatial priors”).

Results: Calibration results showed that light rays between the tissue surface and navigated optical devices were projected with sub-millimeter accuracy. Surface flux measurements were within 5% of diffusion theory predictions. Liquid phantom experiments determined the improvements in quantification of fluorescence yield, with errors of 85% and <20% for no priors and spatial priors, respectively. CBCT-DOT fusion in a VX2-tumor rabbit model delineated contrast enhancement using a dual CT/optical liposomal nanoparticle.

Conclusions: The image-guided computational framework not only enabled co-registration of CBCT and DOT, but moreover improved the performance of fluorescence reconstructions.

Funding Sources: This work was supported by a Canadian Institutes of Health Research Doctoral Award, the Harry Barberian Fund (Otolaryngology-Head & Neck Surgery, University of Toronto) and the Princess Margaret Cancer Foundation.

ABSTRACT. Purpose Prostate Cancer (PCa) continues to be a leading cause of cancer deaths among men in North America. With the introduction of focal therapy, there is hope that the cancer can be treated while maintaining the patient quality of life. PCa MRI-guided focal laser ablation (PCaFLA) is one such ablation technique that has been explored for treating PCa [1]. Existing approaches to PCaFLA are not utilizing computational tools to quantify target coverage either for intra-procedural guidance, or as an objective measure for retrospective evaluation of the ablation margins. Our goal was to apply and evaluate open-source registration tools within 3D Slicer to assess target coverage post-ablation. Methods Three registration techniques were compared for registration of a planning image (defined as same-day MRI examination prior to any ablation) and post-ablation image (defined as same-day MRI examination immediately after completion of ablation) for 10 cases from a previously completed phase 2 clinical trial [2]. The techniques compared are all available via 3D Slicer and are: BRAINSFit [3], Elastix [4], and SegmentRegistration; a deformable registration method using distance maps and B-spline regularization described by Fedorov et al. [5]. For each case, the best technique was selected for registration. The tumor region of interest and ablation zone were outlined manually by a radiologist, and the prostate gland was segmented. Overlap of ablation zone with the tumor volume and target volume (tumor plus a 5mm safety margin) was quantified using the following metrics: (1) percent overlap of tumor volume defined as the percentage of the tumor volume within the ablation zone; (2) percent overlap of target volume defined as the percentage of the target volume within the ablation zone; (3) 95% Hausdorff distance (HD); (4) maximum HD; and (5) average HD. Results Mean tumor and ablation volumes were 0.103 cc and 3.91 cc, respectively. Quantification of the overlap (see fig. 1) between ablation volume and tumor volume revealed partial overlap in most cases (80%), with an average overlap of 39.5%. Mean 95% HD (13.1 mm), maximum HD (24.3 mm), and average HD (7.4 mm) were computed between the ablation volume and tumor volume. Results for ablation volume and target volume overlap will be presented at the time of the meeting. Conclusions We evaluated the feasibility of applying open-source computational tools to MRI-guided focal laser ablation. We identified registration tools and settings that provide satisfactory results for the data evaluated based on qualitative assessment. We also demonstrated the potential for quantitative evaluation of the ablation coverage, which could improve intra-procedural feedback and post-procedural assessment. Future work needs to be done regarding registration evaluation and optimal metrics for evaluating overlap. Funding Source(s) P41 EB015898. References 1. Oto et al. (2013). MR imaging–guided focal laser ablation for prostate cancer: phase I trial. Radiology , 267 (3), 932-940. 2. Eggener et al. (2016). Phase II evaluation of magnetic resonance imaging guided focal laser ablation of prostate cancer. J Urol , 196 (6), 1670-1675. 3. Johnson et al. (2007). BRAINSFit: mutual information rigid registrations of whole-brain 3D images, using the insight toolkit. Insight J , 57 (1). 4. Klein et al. (2010). Elastix: a toolbox for intensity-based medical image registration. IEEE TMI , 29 (1), 196-205. 5. Fedorov et al. (2015). Open-source image registration for MRI–TRUS fusion-guided prostate interventions. IJCARS , 10 (6), 925-934.

ABSTRACT. Purpose Despite the groundbreaking evolution in nerve-sparing prostatectomy, there is still high risk of post-operative complications such as erectile dysfunction. Because of the proximity of the cavernous nerves (CN) to the prostate gland, these nerves are at risk of injury during the removal of cancerous prostate gland. Also, it is still unknown how the complex branches of CN (CNB) contribute to the erectile function of a patient. Therefore, intra-operative image guidance on nerve functionality is needed to maximally prevent the post-operative complications. In this study, we present in vivo proof-of-concept of novel near-infrared fluorescence VSD imaging of nerves on rodent animal model, which can provide real-time, wide field-of-view imaging with short staining time based on direct VSD administration.

Methods For the proof-of-concept study, adult male Sprague-Dawley rats (325-350 g; Charles River Breeding, US) were used. The 200 µl of 1mM VSD (IR780 perchlorate) in DSMO + cremaphore solvent was directly administrated to the exposed rat prostate. After 10 min of staining, the VSD not bound to the tissue was flushed out with 2-ml PBS solution. The customized coherent fiber bundle (Myraid Fiber Imaging, US) comprised with 50K fiber cores was coupled to the sCMOS camera (ORCA Flash 4.0, Hamamatsu K.K., Japan) through a long-pass filter at 800 nm (Fig 1a). For the laser illumination, a 100mW laser diode at 780 nm was used. The bipolar electrical electrode was installed to the right CN. The FL emission was recorded for 5 min, and electrical stimulation was applied for 1 min with 4-volt square-wave pulse for 5 ms duration at 16 Hz in the middle of recording. An intracavernosal pressure (ICP) was recorded in parallel to validate the successful stimulation of erectile function.

Results Histopathological analysis validated the successful VSD staining through the levator fascia whose thickness is few hundreds of µm: The round cross-sections of CNB were successfully differentiated (Fig. 1b). The stimulation on erectile function was also validated with the ICP increase significantly differentiable from the basal level at pre-stimulation phase: 90.98 ± 0.89 mmHg vs. 12.04 mmHg, respectively (Fig. 1c). Fig. 1d presents that the FL images revealed the functional contrast on branching CNB structures with up to 10 % of F/F0 with stimulation, while it was vanished back to basal level in the post-stimulation phase.

Conclusions In this study, we presented in vivo proof-of-concept of novel nerve-guidance method amenable to the current standard in nerve-sparing prostatectomy. We plan to collect more animal cases, and will further advance the experimental design to quantify the contribution of CNB on erectile function.

Funding Source(s) Financial support was provided by the NIH Brain Initiative under Grant No. R24MH106083-03 and the NIH NIBIB under Grant No. R01EB01963. Jeeun Kang, Ph.D. is supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1A6A3A03011551).

ABSTRACT. Roozbeh Shams1, Fabien Picot1,2, Cynthia Menard2, Frederic Leblond1,2, Samuel Kadoury1,2 1 - Polytechnique Montreal, Montreal, Canada 2 – CHUM research center, Montreal, Canada Background Double-sextant biopsies remain the standard of care for prostate cancer diagnosis. This method can lead to 40-45% false negative rate at the initial biopsy. Therefore, a method to accurately confirm biopsy sampling in real time would be beneficial for timely cancer detection. MRI-TRUS fusion to localize tumors has shown improved detection accuracy, but cannot confirm cancerous tissue. Raman spectroscopy is an emerging optical technique that has been shown to be able to discriminate cancerous from healthy tissue, and classify the grade of the tumor. We demonstrate the first system in 3D Slicer towards an electromagnetically tracked Raman spectroscopy device for interstitial prostate procedures, such as HDR brachytherapy, based on diffusion reflectance. Material and Methods We implemented a module extension in 3D Slicer to demonstrate our workflow involving an EM tracked optical sensing needle (Aurora, NDI) with an embedded 6-dof sensor and visualize classified tumor locations. An optical system (MCWHL5, Thorlabs) producing a white light source, a spectrometer (QE65pro, Ocean Optics) and an optical probe (assembled by FiberTech Optica) were used. A 3D printed phantom with cavities was fabricated to test the system. The cavities were filled with dyed Polydimethylsiloxane (PDMS) mixed with TiO2. Red and blue dye were added to change the optical response of PDMS and TiO2 was added to increase the diffusion of the mixture. Also, silicon landmarks were added on the surface for calibration. An SVM classifier with an RBF kernel was trained in Matlab based on prepared dyed PDMS samples. This model was then used to classify the blue PDMS (for cancerous tissue) vs red PDMS (for normal tissue), and communicated with 3D Slicer through OpenIGTLink. Results For the experiments, the phantom was submerged in water and a T1 weighted MRI was acquired. The calibration process consisted of finding the transformation between surface landmarks on the phantom in EM and in MRI coordinate spaces. Moreover, the optical probe was tracked and used to navigate towards pre-identified locations on the MRI and confirm the detected “cancerous tissue”. With the current setup, a registration error of 2.71mm was achieved between MR and tracking coordinate systems. The result of the visualization can be seen in Figure 1. All of the 8 embedded dyed PDMS markers were successfully detected, as shown in Figure 1 with blue spheres. The red points display the location of the tracked optical probe. Conclusion An image navigation system developed in 3D Slicer using an EM tracked optical sensing needle was demonstrated. Future work will integrate TRUS/MRI fusion, improving navigation accuracy.

ABSTRACT. Purpose: Medical imaging plays an important role in various diagnosis, pre-/intra- procedural interventions, and post analysis. Advanced imaging and image processing technologies have enabled superior visualization and localization allowing accurate diagnosis and treatment guidance. For instance, 3D Slicer (www.slicer.org) has been used in numerous medical studies and clinical procedures that can only be possible with the aid of the visualization software. Image handling algorithms and other functions have greatly increased the use of advanced visualization software. As such, user-controlled image navigation plays a vital role in revealing a particular view from the information-rich image set i.e. locating the most informative sectional view(s) of a 3D target anatomy constructed from a series of DICOM images. 3D Slicer developments include segmentation algorithm, surgical planning and navigation, and atlas generation among many others. Despite robust volumetric image handling, user interface and resulting view from such tasks are projected on a 2D screen, requiring the user to map the 2D perception to the true 3D anatomy that is inside the human body. In terms of physical input and output, a 2D screen and PC mouse have been used as common tools to interactively manipulate images. As alternatives, virtual and/or augmented reality have been investigated to provide greater 3D perception. However, such devices often require goggle-type wearable setup that restricts physicians’ regular vision and activity. Most of these technologies are purely virtual as they do not engage the user physically but provide visual information only. The user then needs to map this information to the real site. Other newer technologies such as 3D hologram and spatial input devices have also been investigated for medical use. However, the combination of a 2D display i.e. monitor and a planar view control i.e. PC mouse dominates in practice similar to the radiology reading. This leads us to question if a fixed monitor with PC mouse is the optimal visualization method particularly in medicine. Methods: Regarding coordinate system, visual information given by medical image, visualization software and virtual reality setup are commonly based on a fixed object coordinate system e.g. RAS coordinate in DICOM. The users then control viewer location i.e. pan, rotate, and zoom to visualize a 2D section view with some level of 3D visualization aid since a fixed 3D view can be confusing without manipulating. This process typically includes multiple steps of direct visual perception and mapping of images between 2D and 3D spaces. The purpose of this study is to develop and evaluate an intuitive image control and viewing device for 3D visualization of volumetric medical images that can coexist in the physical object coordinate system with a simple Cartesian offset so that the user can physically control viewing next to the real anatomy without coordinate transformation i.e. mapping, which could enable instant perception and utilization of advanced medical imaging technologies. As a concept development, we designed “Air Slicer”, a visualization aid device for image-based diagnosis and therapy guidance. Results: We designed a prototype system that can be used in typical imaging and operating facilities. The system consists of a passive robotic arm that can hold a conventional or transparent display monitor with mechanical counterbalancing. A real-time cross-sectional image can be continuously displayed by tracking kinematic parameters of each joint of the arm. The arm can be attached to an operating table, imaging devices i.e. MRI, CT, or a standalone table. Conclusions: The proposed Air Slicer can provide more intuitive image handling over the conventional image viewers that employ a typical input and output method (i.e. PC mouse and monitor). Using Air Slicer, the user could instantly articulate slicing location and angle to obtain a complex cross-sectional view using true spatial perception and real scale physical control. Funding Source: In search of funding to prototype and evaluate the disruptive approach.

ABSTRACT. Radiomic Analysis of Peritumoral Tissue on Contrast Enhanced Breast MRI Prior to Neoadjuvant Chemotherapy of Breast Cancer Carlos Aramayo, Eva Gombos MD, Jayender Jagadeesan PhD Brigham and Women’s Hospital Department of Radiology

Purpose: To identify key radiomic features in pretreatment peritumoral breast tissue associated with high response to neoadjuvant chemotherapy.

Methods: In this retrospective study, we identified 27 biopsy-proven invasive breast cancer cases (BIRADS 6) with dynamic contrast-enhanced MR (DCE-MRI) studies performed prior to neoadjuvant chemotherapy. Segmentation of pretreatment MR images was completed using the open-access image-processing software 3D Slicer. For each case, a tumor label map and five distinct masks were created by thresholding subtracted pre and 1st post-contrast images and dilating the tumor label map to voxel-equivalent distances of 2.5mm, 5mm, 10mm, 15mm and 20mm beyond the tumor border.

After segmentation, a total of 57 metrics including distribution statistics, shape, morphology and texture were computed from each tumor label map and peritumoral mask using the 3D Slicer HeterogeneityCAD module. To assess response to treatment, cases were separated into 2 groups using Miller-Payne scores acquired from final surgical pathology reports. Scores of 1-3 were placed into a “low-response” group while scores of 4-5, indicating reduction of tumor cells greater than 90%, were placed into a “high-response” group. Statistical correlation was performed using univariate Mann-Whitney tests to compare low-response and high-response groups at the tumor and individual peritumoral levels.

Results: Seven metrics showed significant difference between the low-response group and the high-response group at the tumor level, 2.5 mm peritumoral level, and 5mm peritumoral level. Among these metrics were Volume (in mm^3), and Maximum 3D Diameter (P≤0.05). Metrics for Sphericity, Long Run Enhancement (indicative of coarseness), Surface Area (in mm^2), and Compactness 2 achieved a statistically significant difference of P≤0.05 at the tumor level and 5mm peritumoral level, and of P≤0.001 at the 2.5mm peritumoral level.

Conclusions: Our study showed that radiomic features of peritumoral breast tissue extracted from DCE-MRI, such as lower coarseness, and greater volume, surface area, sphericity, and compactness are associated with a higher response to neoadjuvant chemotherapies.

Funding Source(s) National Institutes of Health through Grant Numbers P41EB015898 and P41RR019703

ABSTRACT. Purpose The supplementary motor area (SMA) and the pre-supplementary motor area (pre-SMA) are critical regions within the brain’s language network. For patients undergoing tumor resection, an important step in preoperative planning is to identify these areas. While the available literature classifies the importance of the SMA and pre-SMA and their connectivity, reports assessing the pertinence of these features in the context of brain tumors are limited. The purpose of this study is to discuss the SMA and pre-SMA connectivity using Unscented Kalman Filter (UKF) tractography in patients with frontal lobe lesions.

Methods Forty patients with left frontal tumors were evaluated using the UKF method. Structural connectivity was assessed using seed-based tractography and functional connectivity was assessed using resting-state fMRI.

Results The connectivity pattern of the SMA and pre-SMA was highly variable, however insightful when correlated to the pre and post-operative clinical outcome.

Conclusion UKF tractography is an effective method of visualizing the white matter tracts that connect the SMA and pre-SMA into the language network in patients with brain lesions. Such visualization can inform neurosurgeons during preoperative planning about the structural and functional connectivity of these eloquent cortex areas.

ABSTRACT. Purpose: In this study, we aim to assess the feasibility of using a movie-watching functional magnetic resonance imaging (fMRI) paradigm to extract language network in brain tumor patients. The language tasks used in the conventional task-based fMRI can be difficult for patients to perform, particularly for patients with language or neurological deficits. This limitation may result in decreased detection power of task-based fMRI. We propose a less-demanding, "task-free" condition where patients only need to passively watch a movie clip that contains language feature. Methods: Nine brain tumor patients (2 males, 7 females, mean age = 43.8 years, range: 28-59 years, 7 right-handed, 2 left-handed) participated this study. These patients had brain tumors located in the left frontal or temporal/parietal regions, and four of them expressed pre-operative language deficits. During movie-watching condition, the patients were shown a 7-min clip that contained dialogues and non-speech segments. They also underwent a vocalized antonym generation task-based fMRI. A language response model (LRM) derived based on temporal response of a separate group of healthy subjects' language area (who also watched the same movie clip) was applied as a linear regressor in the first-level general linear model (GLM) analyses of the patients' movie-watching data, to identify brain regions whose time courses were highly correlated with the LRM. The resulting statistical maps of the movie-watching fMRI were compared with that of the task-based fMRI to assess the effectiveness of mapping language areas in brain tumor patients with the movie-watching fMRI. Results: In general, when comparing the language maps at the same statistical level, the movie-watching fMRI identified more activated voxels in the bilateral temporal language region, while the task-based fMRI revealed more frontal area activations. Activation location in the putative language regions of the two paradigms were partially overlapped, in particular, the activations in Broca's area were quite distinctly located. In addition, when the task-based fMRI failed to reveal left frontal activations, the movie-watching fMRI successfully localized Broca’s area in two patients. Furthermore, the movie-watching maps were more specific to language function, while the task-based maps were less language-specific with extraneous activations in non-language regions. Conclusions: This initial investigation of the novel fMRI paradigm suggests that it is feasible to map language function in brain tumor patients with a "task-free" movie-watching condition. This paradigm may be more effective to demonstrate the extensive neural substrates for natural language processing, which may not be completely revealed using traditional language tasks with isolated and controlled conditions. Further research should be focused on optimization of analytic approach and validation in neurosurgical patients against clinical gold-standard intraoperative language mapping techniques.

ABSTRACT. Walid Abdelmoula1, Begona Lopez1, Michael Regan1, Elizabeth Randall2, Sean Lawler1, Jann Sarkaria3, Jeffrey N. Agar4, Tina Kapur2, William Wells2, Nathalie Y.R. Agar1,2 1Department of Neurosurgery, 2Department of Radiology, Brigham and Women’s Hospital, Boston, USA; 3Radiation Oncology, Mayo Clinic; 4Department of Chemistry and Chemical Biology, Northeastern University Purpose: Mass spectrometry imaging (MSI) provides a wide-range of targeted and untargeted spatially-resolved biomolecular information; making it a promising discovery tool for many biological applications. Multi-modal integration between MSI and MRI data would enable cross correlating molecular and anatomical phenotypes providing enhanced biological and diagnostic insight. Automatic co-registration between MSI/MRI data is a computationally challenging process due to: 1) dimensional complexity that allows one-to-many mapping (hyperspectral vs. anatomical images); 2) lack of spatial correspondences; and 3) elastic deformation from MSI tissue processing. In addition, identification of informative molecular patterns is limited by large size, sparsity, and non-linearity of high dimensional spectral data. We propose a computational platform using stochastic neighbor embedding to overcome these challenges. Methods: A patient derived xenograft (PDX) mouse brain model of glioblastoma was first scanned using in vivo 7 Tesla MRI, and the dissected brain was then sectioned and processed for 3D MALDI 9.4Tesla MSI. The non-linear dimensionality reduction method of t-distributed stochastic neighbor embedding (t-SNE) was used to project the high dimensional data-points (spectra) into a 3D space based on the pairwise molecular similarities while preserving spectral local structures. The t-SNE features were spatially organized forming a t-SNE image that revealed molecularly distinct regions and thus establishing spatial correspondences with the MR image through image registration[1]. The non-linear transformation model of B-Spline was used to capture the local deformation, and the registration was implemented in a multi-resolution manner. Data-driven identification of multi-scale molecular patterns was accomplished using recent developments based on hierarchical stochastic neighbor embedding (HSNE)[2]. Those molecular patterns were correlated with the original spectra to reveal the highly co-localized molecular features. Results: 3D MSI data was non-linearly aligned to MRI. The parent ion of the EGFR inhibitor erlotinib was identified at mass-to-charge ratio (m/z) 394.1756 and mapped to the MR image as shown in Figure 1; revealing elevation of drug signal within the tumor region. Two molecular patterns associated with tumor and normal regions were identified by HSNE, and molecular ions contributing to these patterns were revealed and mapped to the MRI volume. Conclusions: We have developed a computational platform to: 1) non-linearly align MSI-MRI data volumes, and 2) identify molecular patterns from complex omics data. Funding Source: NIH U54 CA210180 MIT/Mayo Physical Science Oncology Center for Drug Distribution and Efficacy in Brain Tumors, and the DFCI PLGA Fund; NIH R25 to E.C.R. (R25 CA-89017) in partnership with the Ferenc Jolesz National Centre for Image Guided Therapy at BWH. References: [1] W. M. Abdelmoula et al., “Automatic generic registration of mass spectrometry imaging data to histology using nonlinear stochastic embedding,” Anal Chem, vol. 86, no. 18, pp. 9204–9211, 2014. [2] W. M. Abdelmoula et al., “Interactive Visual Exploration of 3D Mass Spectrometry Imaging Data Using Hierarchical Stochastic Neighbor Embedding Reveals Spatiomolecular Structures at Full Data Resolution,” J. Proteome Res., 2018.

ABSTRACT. Purpose: The intra-arterial (IA) route provides local, highly selective therapeutic agent delivery to the brain while minimizing systemic toxicity. The blood brain barrier (BBB), however, is a major obstacle preventing effective transfer from vascular to parenchymal compartments. Opening of the BBB using hyperosmotic mannitol has been used in clinical applications for several decades, but high variability in BBB opening (BBBO) translated to inconsistent clinical outcomes and resultant limited acceptance of this technique. Recent progress in MRI hardware and software development dramatically improved the speed of acquisition with excellent signal to noise, improving the prospects for performing endovascular neurointerventions. We postulate that the precision of BBBO could be greatly improved using its real-time, dynamic visualization.

Methods: We utilized an ultra-fast MR imaging method that combines a short-echo time radial FLASH MRI sequence with high undersampling and in-line image reconstruction by parallelized nonlinear inversion with regularization implemented on multiple GPUs. The resulting serial images had an acquisition time of 50ms per slice with strong T1 weighting, allowing dynamic imaging of the entire canine brain (20 slices) with high temporal resolution of 1Hz/volume. Two dogs were subjected to either middle cerebral artery or basilar artery catheterization under fluoroscopic guidance. The dogs were then transferred to a 3T MRI scanner (Siemens Prisma) and 25% IA mannitol mixed with gadovist at the ratio 1:50 (Mn+Gd) was infused over one minute while acquiring T1 weighted MR images in real-time. Each dog was subjected to three procedures over the course of several weeks.

Results: Dynamic MRI detected T1 enhancement immediately when BBB opened, at very high temporal resolution of 1Hz. As shown in Fig.1 T1 enhancement was detected in the distinct brain region supplied by the microcatheter. There was gradual increase in signal intensity shown between frames 98 and 177. The area shaded in blue represents the period of Mn+Gd infusion. BBBO was observed 30-40s from the beginning of infusion. The area of BBB opening was highly restricted to the perfusion area supplied by the microcatheter. Once BBBO was visualized, it was possible to immediately reposition the microcatheter and open the BBB in adjacent regions. Conclusions: We found that dynamic MR imaging is excellent for visualizing the process of BBBO. We believe this approach will improve the efficacy and safety of IA drug delivery to the brain.

Funding Sources: NIH/NINDS R21NS091599; R01NS091110; R01 EB007829

ABSTRACT. Purpose: The blood-brain barrier (BBB) is a primary obstacle inhibiting effective drug delivery to brain tumors. Intra-arterial (IA) mannitol infusion is known to incite BBB permeability, however, the reproducibility of BBB opening (BBBO) has been low and therapeutic outcomes vary. We previously showed in large animal models that the variability can be reduced or even eliminated when BBBO is performed under real-time MRI. Here we optimized this approach for a mouse glioblastoma model enabling high-throughput, cost-effective efficacy studies. Methods: BBBO: Scid mice (20-25) were anesthetized with 2% isoflurane. Mice were catheterized with a microcatheter in the common carotid artery and imaged with a Bruker 11.7T horizontal bore MRI scanner. Baseline T1- and T2-w and dynamic GE-EPI images were acquired. IA Feridex (0.1 mg Fe/ml) was infused under dynamic GE-EPI to predict the perfusion territory for different injection speeds. 25% mannitol was delivered IA over one minute at the optimal speed determined by the prior Feridex perfusion study for each animal. MRI and histology were used to assess the BBB status and any consequences compromising procedural safety. IA chemotherapy: Mice were inoculated with luciferase-expressing human glioblastoma (GBM1) cells. Three weeks after glioblastoma induction, mice were subjected to BBBO as described above and melphalan, (0.05 mg/mouse) was injected IA. Mice were serially imaged with bioluminescence imaging (BLI) to monitor tumor growth. Results Dynamic susceptibility contrast MRI during transcatheter IA infusion of SPIO contrast demonstrated that at a rate below 0.1 ml/min, cerebral perfusion was inconsistent. However, when the speed was increased to 0.15 ml/min, the desired brain perfusion was obtained as visualized by the reduction in signal intensity during injection of the Feridex bolus. IA mannitol infused at the pre-determined rate over 60 s resulted in BBBO as verified by gadolinium-enhanced T1 weighted MRI, showing hyperintensity in the region previously highlighted by the Feridex infusion. The SPIO perfusion MRI showed an average signal change area of 26.00±5.60%, while Gd-enhanced MRI showed an average signal change area of 26.52±5.33%, both measures were in good agreement (r=0.879). In histopathology, IV-injected Evans Blue (marker for BBBO) and rhodamine (surrogate marker of therapeutic agent) demonstrated a pattern of extravasation that was consistent with that observed by MRI. T2-w MRI acquired on day 3 post BBBO showed no obvious abnormalities and no Gd-enhancement on MRI. Immunohistochemistry showed no increased number of GFAP+ activated astrocytes or IBA-1+ microglia, nor loss of NeuN+ neurons. Altogether, these data indicate an excellent safety profile. The tumor-bearing mice receiving melphalan treatment with or without BBBO showed a gradual BLI signal increase, which after six days, decreased in the group with BBBO indicating a treatment response. This signal reduction was transient, suggesting that a single dose does not completely arrest tumor growth. Conclusions Real-time MRI-guidance enables precise, safe and effective local BBBO with high reproducibility. Bypassing the BBB resulted in transient treatment response, indicating the need for additional studies to achieve a more sustained treatment effect. Funding Source: NIH/NINDS R01NS091110

ABSTRACT. Purpose Magnetic resonance imaging guided focused ultrasound (MRgFUS) ablation is an emerging technique for noninvasive transcranial surgery. One of the major issues of traditional transcranial MRgFUS ablation is that the pressure and time-averaged power required to generate a lesion is relatively high which prohibits the use at peripheral region of the brain. Cavitation nucleation agents like phase shift nanoemulsions (PSNE) have been shown to reduce the acoustic pressure required to generate lesions through inertial cavitation and thus the treatment envelope could be broadened. In this study, we investigated an emerging type of PSNE made from perfluorobutane (PFB) that could facilitate mechanical ablation in the brain by acoustic droplet vaporization at relatively low pressures (i.e. <= 2.0 MPa). The success of this project would provide a useful tool to enable transcranial ablation. Methods The PFB PSNE were synthesized by compressing (~45 psi) the microbubbles under dry ice bath (-10~-15 oC). Acoustic emissions from vaporizations of the droplets were recorded in vivo in the brain in SD Rat (N = 3) using a passive acoustic detector that is sensitive to 1.5 MHz. Sonication was performed using a HIFU transducer operating at 690 kHz. For each animal, transducer focus location and rat brain were first visualized using T1-weighted MRI and the transducer focus were moved to the target region. Sonications were targeted on the striatum on both hemispheres (left hemisphere: sonication without PSNE; right hemisphere: sonication 1 min after i.v. PSNE injection through the tail vein followed by a saline flush). In each sonication, we started with a ramping pressure pulse (pressure ranging from 0.9 to 2.0 MPa) followed by a steady pressure pulse (2.0 MPa) for 10 mins. Animals were euthanized 24 hours after experiments. The brains were harvested after cardiac perfusion, and tissue damage was confirmed with H&E staining. Results Vaporization was confirmed by acoustic emission. Contrast enhanced T1-weighted MR images showed an increase of intensity at the targeted region. The onset of the acoustic emission occurred at 1.3~1.8 MPa. Its amplitude then elevated steeply as the exposure level increased and then saturated. The histology results showed evidence of nonthermal ablation. A well-defined region was found consisting of neuron karyopyknosis. Red blood cell extravasation was obvious around large blood vessels, presumably the result of droplet vaporization and subsequent inertial cavitation that ruptured the vessel. Conclusions The results suggested that we could successfully induce PFB PSNE vaporization in intact rat brain with tolerable pressure using a single focused ultrasound transducer system. The corresponding damage was confined and had a clear boundary which provided us with a promising way to focally ablate brain tissue using HIFU. Future work will examine these agents in brain tumor models. Funding Sources This work is supported by NIH grant P01CA17464501 and BU-BWH partnership.

ABSTRACT. Purpose To develop a method of fusing pre/intra-operative Computed Tomography (CT) with pre-operative Magnetic Resonance Imaging (MRI) and evaluate the impact of using the fused data on the implantation of iodine-125 (125I) seeds for brachytherapy of recurrent gliomas. Methods A study was performed on a cohort of 19 consecutive patients with recurrent gliomas were treated by 125I brachytherapy with CT-MRI fusion image guided (CMGB), and 23 patients treated with CT alone guided (CGB). After IRB approval, all patients with recurrent gliomas after failing prior surgical resection, whole brain radiation therapy and/or stereotactic radiosurgery were accrued from 2007 to 2017. Statistical analysis was performed to compare (1) the planning target volume, (2) the accuracy of location of catheters, (3) the target volume covered by 150% prescribe dose (V150), (4) the target volume covered by 200% prescribe dose (V200), and (5) the conformity index (CI) with or without fused data. Results The median planning target volume was 34.4 cm3 (range, 8.8-137.4 cm3) in CGB, and 46.1 cm3 (range, 10.5-87 cm3) in CMGB with significant difference (p < 0.05). The accuracy of catheter insertion was 93.2% with CMGB and 79.3% with CGB. The median V150 and V200 was 70.45% vs 73.9% and 54.24% vs 40.20% in CGB and CMGB, respectively. There was significant difference for CI (82.6% vs. 73.9%, p < 0.05) in the two groups for the post-operative verification. Conclusions The proposed MRI-CT fusion method enables a quantitative assessment of impact on recurrent gliomas with brachytherapy. The additional information obtained from the fused images can be utilized for more accurate delineation of lesion boundaries and targeting of catheters. Experimental results show that the image fusion algorithm is robust and reliable in clinical practice.

ABSTRACT. Purpose: Thermal ablations are standard therapeutic options to destroy pathological tissues in many organs such as liver, kidney, lungs, etc. Clinicians aim at an ablation that is slightly larger than the target to reduce the chance of recurrences, while minimizing damage to nearby healthy and vulnerable tissue. For those reasons, thermal monitoring is instrumental to achieve a precise ablation. It can be done in real-time, without ionizing radiation and at a low-cost by ultrasound. Methods: We propose to leverage ablation therapy systems by integrating an ultrasound element to the ablation probe. The key concept of this technology is the communication between this integrated ultrasound element and any diagnostic ultrasound probe. To estimate the temperature of the ablation region, we exploit the temperature dependence of the tissue speed of sound (SoS). The ultrasound element transmits ultrasound waves carrying time-of-flight information that are affected by the temperature changes during ablation. We use directly the time-of-flight (TOF) information to solve a limited angle US tomography reconstruction problem, that we regularized by a physics-based simulation model. Results Promising preliminary results have been obtained with this approach to monitor HIFU and different ablation patterns generated by a novel bipolar radiofrequency ablation system. Conclusions: We have shown that intra-operative ultrasound information coupled to thermal simulations has the potential to monitor different thermal ablation procedures such as HIFU and RFA in real-time while ensuring clinically acceptable accuracy and precision of the treatment. The successful implementation of this technology would allow to substantially reduce the therapy cost compared to MR-based temperature monitoring technique.

ABSTRACT. L.A.Groves1, A.Rankin1, T.Peters1, E.C.S.Chen1 1Robarts Research Institute, School of Biomedical Engineering, Western University

Purpose: Minimally invasive procedures are becoming increasingly more reliant on ultrasound (US) guidance. Often these systems incorporate an augmented reality (AR) environment for improved navigation through providing the surgeon with medical images and virtual models in a single field of view. A fundamental requirement to integrate US images into AR environments is US probe calibration, which places images from a tracked US probe in the context of the spatial tracker, providing a common frame of reference for tracked surgical tools and images. The current US probe calibration methods are limited, as many require specific phantoms, large number of images, knowledge of US physics, time-consuming processes and few are publically available. Despite the prevalence of many US calibration methods, there is still a need for a simple and effective calibration method that users of any level can perform. The method published by Chen et al. (GUSCAL) was implemented as a 3D Slicer module. The focus of this research is to test the robustness and usability of the method as a 3D Slicer module in a novice user study.

Methods: The US calibration is formulated as a point-line registration between homologous datasets; the user is required to insert a tracked needle (line) into the US beam, causing a hyperechoic reflection (point). As a registration problem, the accuracy of US calibration is influenced, among other things, by the location of the US reflections and needle orientations. The US reflection is segmented manually, requiring users to freeze the US and tracking stream, and select the needle centroid. Users repeated this process 14 times at various needle locations and orientations, performing the calibration under three different conditions: (1) free-hand with one focal depth, (2) free-hand with three focal depths, and (3) using a mechanical arm to fix the needle with one focal depth. All other settings on the US scanner remained consistent and a randomized order was assigned to each subject. The discrepancies between different user’s calibrations per condition were compared to gain insight into the robustness and usability of the algorithm.

Preliminary Results: The current user population is n=15. A gold standard transformation matrix was obtained by an expert user performing a rigorous calibration. The users’ results will be compared to this gold standard through performing point reconstruction with the image corners as targets for all final transformation matrices per condition. The users average distance from the gold standards reconstructed locations and the average time are presented.

Note: table and figure included in uploaded submission

Conclusions: The results presented in Table 1, suggest that condition 1 is both the most true and precise. This condition consistently has the lowest average and standard deviations, as well as the lowest time. However, in general the average and standard deviation values are low suggesting good overall robustness. The low completion time for novice users is further indicative to the usability of this implementation. To achieve a sufficient calibration effectively performing it free-hand with one focal depth is recommended for any user. The GUSCAL, a 3D Slicer module for ultrasound calibration, will soon to be publicly available as an open-source software. Funding Source(s): CIHR Foundation Grant and Canadian Foundation for Innovation.

ABSTRACT. Purpose: With increasing use of minimally invasive instruments, such as steerable catheters or biopsy needles, ultrasound grayscale (B-mode) imaging emerged as a portable, nonionizing, and broadly available method for spatial guidance. However, imaging noise and scattering, as well as motion outside of the scanning field or the inability to distinguish an instrument from its anatomical surroundings impede B-mode guidance. The purpose of our work is to complement B-mode imaging with an imaging modality, which is routinely available within the same clinical ultrasound system and will improve reliability of navigation of minimally invasive instruments.

Methods: We use color Doppler flow imaging in a new application – for image guidance. A piezoelectric crystal is affixed to a minimally invasive instrument. The crystal is driven by a square-wave signal with frequency typically in the range of 90 kHz to 110 kHz and amplitude within a range from hundreds of millivolts to single volts. The continuous square-wave signal produces high-order harmonics that are received and interpreted by the ultrasound system as a Doppler shift. This apparent Doppler shift produces an instantaneous elongated color marker along a Doppler beam that intersects the location of the vibrating crystal.

Results: We have equipped various catheters, needles, and cannulas with piezoelectric crystals ranging from 0.5 mm to 3 mm in diameter and guided them into abdominal and cardiovascular anatomical targets in anesthetized pigs. As one example, Figure 1 shows a prototype of a transcutaneous biopsy needle. The crystal is attached to the tip of a stylus, which is inside the needle lumen during insertion. Figure 2 shows an instantaneous color Doppler marker that is superimposed on the B-mode scan and guides the needle tip into pig’s kidney to mimic biopsy conditions. More than one crystal can be a source of the guidance signal and marker colors can be made different by adjusting the driving signal frequency. Blood flow visualization in the heart and vessels can be removed by reducing Doppler gain so that only the color marker is depicted.

Conclusions: We present a new concept of using color Doppler imaging as a method for guidance of customized minimally invasive instruments. The instantaneous color Doppler marker complements B-mode ultrasound image guidance by minimizing or eliminating uncertainty in identification and real time tracking of the minimally invasive instruments in abdominal and cardiovascular applications. More than one instrument can be tracked at a time.

Funding Source: NIH Grant R01EB019947 (M.B. and M.F.)

ABSTRACT. Purpose Real-time imaging guidance during radiation therapy procedures today require new specialized combined MRI-LINAC systems (1-2). This project tests the feasibility of using simultaneous MRI and volumetric ultrasound, but acquired during pre-treatment, to provide real-time guidance during radiation therapy in existing LINAC systems, i.e., providing the soft-tissue contrast of MRI for guidance without requiring a combined MRI-LINAC system. Methods Rather than registering static or multi-phase MRI images to real-time ultrasound images during therapy, our approach utilizes MR and ultrasound images that are acquired simultaneously during pre-treatment. During radiation treatment in a LINAC, only ultrasound is used to determine the current respiratory state. By matching the respiratory state at the current time to equivalent respiratory state images acquired during pre-treatment, virtual real-time MRI can be used to provide the necessary soft-tissue contrast for image guidance. A prerequisite for accomplishing the image guidance is to provide hands-free, electronically steerable 4D ultrasound images that adequately covers the tumor target and also adjacent endogeneous fiducial markers simultaneously with MRI. In addition, during the LINAC therapy session, volumetric 4D ultrasound re-acquires the fiducial markers to determine the respiratory state, and match the current images with the pre-treatment images via look-up. This required the development of an MRI and X-ray compatible 4D ultrasound probe and also fast feature matching algorithms to display the correct pre-acquired MR images in real-time. An 18,000 element e4D ultrasound probe was built with an extended 8.5 m cable that was triple-shielded to avoid interactions with the MRI system. All MR-incompatible components were removed and replaced. Harmonic imaging with 1.7/3.4 MHz was used. The probe was then tested using fast gradient recalled echo (FSPGR) as well as fast-spin echo (FSE) pulse sequences, and also 8- and 32-channel phased array coils in healthy volunteers. Results Simultaneous MR and ultrasound images were successfully acquired without significant image artifacts. Multiphase 2D-FSPGR images with 1.3x1.9x10.0 mm3 spatial resolution were acquired with a temporal resolution of 4 fps (TR=3.3 ms), comparable to the 4-6 fps frame rate of the 4D ultrasound images. The probe was able to easily visualize the liver to a depth of 12-14 cm with a 60o x 30o FOV. Susceptibility artifacts from the ultrasound probe were not entirely eliminated but were well-contained within 1-2 cm of the surface. In observation of the displacement of fiducial markers, respiratory drift was observed. As noted in navigator MRI studies, this drift changes the absolute position of the tumor target relative to the respiratory phase. Strategies to account for this respiratory drift are under development. Conclusions The ability to acquire MR and ultrasound images simultaneously has been demonstrated. Using fast block matching algorithms (3), the displacement of multiple fiducial markers can be determined in real-time. The next step in the project is to match respiratory states as determined by the absolute displacement or position of the fiducial markers such that the planning treatment volume (PTV) margins during radiation therapy can be reduced. Funding sources: NIH R01CA190298. References 1. Park JM. Radiother Oncol 2016; 20:279–85. 2. Raaymakers BW. PMB 2017; 62: L41-50. 3. Shepard AJ. Med Phys 2017; doi:10.1002/mp.12574.

ABSTRACT. Purpose: Image-guided thermal therapy is a clinically accepted procedure for various kinds of cancer treatments. The underlying principle of this therapy includes damaging of the cancer tissues by applying thermal energy; subsequently raising the temperature of the target cancer tissue above a certain threshold. In addition to guiding the therapy, one of the key challenges in such procedure is monitoring the temperature across the region of thermal damage to correctly raise the temperature of the cancer tissue as well as keep the temperature of neighboring healthy tissues below a safe limit to protect them from unwanted damage. Currently, MRI-based approach has been considered as standard for non-invasive temperature monitoring. However, the key limitations are very high cost of MRI, lack of portability and design of custom therapy. Ultrasound (US) has been suggested as an alternative for monitoring the tissue temperature due to its temperature dependent acoustic properties (e.g., speed of sound and attenuation). US has many advantages including its low cost compared to MRI, easy accessibility and easy portability. However, the key challenge is interpreting the non-linear behavior of the US signal with respect to the temperature variation. To address this challenge, we present an US-based approach here exploiting the capability of neural network-based techniques to model such non-linear relationship.

Methods: The proposed approach is based on the communication between an US element near the heating region and a diagnostic US probe held at outside. By injecting a pattern using the US element that is subsequently captured by the US probe as B-mode images, we can extract the temperature information from the pattern changes. For accurate temperature estimation, a deep neural network-based approach is adopted in this work that takes a set of B-mode images (with pattern injected) as inputs and subsequently provides corresponding temperature maps as outputs. In an initial feasibility study, we train and test the developed network using a set of simulation experiments.

Results: A comparison of our predicted temperature with the actual temperature indicates a close fit of our estimation with the reference one.

Conclusions: The promising result in this feasibility study demonstrates the potential of the proposed approach to find applications in image-guided thermal therapy procedures.

ABSTRACT. Purpose Minimizing exposure of healthy tissue to ionizing radiation is the ultimate goal of any Radiation Oncology treatment. However, while IGRT efficacy has clearly been demonstrated, the next step is using non-ionizing imaging to confirm the margins. Ferenz Jolesz was the pioneer in utilizing MRI for registration and localization during external beam treatments culminating in the AMIGO suite. This author has long sought to build on Dr. Jolesz legacy as an innovator and mentor through planning of interventional suites using optical imaging, ultrasound and particularly magnetic resonance.

Methods As an International Health Planner, the author has had the opportunity to promote non-ionizing external beam treatment facilities around the world, from the 2014 master plan for DF/BWHCC, to NGNA-Princess Norah CC in KSA, to CCD/MD Anderson Hongquao, to Peking Union Translational Bldg and finally to the Sunnybrook Health Science Centre in Toronto. Results Of all of the facilities planned and constructed for external beam treatments, the Sunnybrook Odette Cancer Centre MRI-Linac is the first clinical use facility in North America to be constructed with expected certification for patient use in 2019. This presentation will outline the planning efforts at each of these facilities but highlight the impact the novel treatment of MRI guided photon / electron therapy has on the patient.

Conclusions While patient treatments won’t begin until 2019, clinical trials have clearly shown that pervasive and concurrent imaging greatly enhances the efficacy of external beam imaging. Going forward advanced targeted therapies combined with molecular imaging will improve treatments even further

Funding Source(s) While Sunnybrook is one of less than 20 Canadian health facilities to receive NIH funding, the primary source of funds for the MRI-Linac was through the Canadian Institute of Health Research.