ABSTRACT. Introduction: Instrumented load sensors have been used in soft tissue calibration and load distribution in the course of total knee arthroplasty for several years [1-5]. A common criticism points to the inert nature of the soft tissue envelope under anesthesia. This study aims to demonstrate the quantitative changes in the load magnitude and relative distribution as patients raise their leg during a wake-up test under spinal anesthesia.

Methods: This study is a pilot study of 25 patients who performed a volitional straight leg raising maneuver under verbal command at the conclusion of instrumentation of a total knee arthroplasty. The spinal anesthetic formula was modified to allow return of motor function within an hour of the induction. The sensory block and sedation were not affected by this. Video capture of the event was made through a lateral perspective to capture the efficiency: height, knee flexion, duration and speed of the maneuver. The tibial load magnitude and relative compartmental distribution were recorded throughout as seen in Fig 1.

Results: There was found to be variability in the efficiency of the spinal anesthetic recipe. Smaller patients had residual motor block and either could not perform the task satisfactorily or at all. Of the 17 patients who had enough power, the relative increase in magnitude of tibial load force was in the range of 2.8 to 4.3, with an average of 3.4. The relative medial to lateral load distribution was preserved in 15 cases. There were no complications. No patient experienced any negative memory recall from the test.

Discussion: As intra-operative load sensors gain popularity in soft tissue calibration, it is important to determine if the relatively low passive forces measured at surgery predict the forces that would be encountered in weight-bearing (the boundaries of their usefulness) [6]. There is still no consensus on the ideal alignment [7], or force target zone [8-10]. It has been established however that the surgeon’s ability to quantify loads is poor [11]. This study demonstrates that a well- balanced knee can be expected to have a predictable 3 to 4-fold increase in load across the joint under reasonable isometric muscular contraction, while preserving the medial-lateral load symmetry. The ability to preserve this compartmental load symmetry indicates that balanced load values obtained at surgery could be a predictor of clinical success.

ABSTRACT. Introduction The importance of accurate cup positioning during total hip arthroplasty (THA) to minimise post-operative complications has been well defined. However it remains unclear if following navigated THA there is a relationship between the active range of movement, the inclination and anteversion of the implant, and the theoretical range of movement as measured intra-operatively using imageless navigation. Methods Fifteen male patients undergoing primary THA using the Orthopilot (Aesculap AG, Tuttlingen, Germany) imageless navigation system were recruited. The Orthopilot system flexion value was recorded. Three months post-THA patients underwent 3D biomechanical analysis (Vicon Motion Systems, Oxford, UK) during which patients performed a standing active hip flexion movement and a sitting task. Inclination and anteversion calculated according to Pradhan’s formula were taken from post-operative radiographs. Findings There was no observed correlation with the theoretical hip flexion and the standing peak active or sitting peak hip flexion values in this series. However, Orthopilot flexion considers only the interaction of the implant components and not the soft tissues surrounding the hip joint which may limit clinical flexion. There does not appear to be any relationship between the inclination and anteversion angles and the range of movement observed post-operatively during a standing active hip flexion task and sitting task. It was observed that peak active hip flexion between the operated and non-operated limbs was not significantly different at three months, which indicates a good level of symmetry at this time point.

ABSTRACT. Background Prior research has not evaluated how femoral head coverage changes as the pelvic sagittal inclination (PSI) is altered. In this computational simulation study, the effect of PSI on 3-D head coverage was quantified and the relationship between regional three-dimensional head coverage and PSI was analyzed.

Methods Eleven control participants (6 male) were enrolled. Surface models of the pelvis and femur were generated from computed tomography images. The PSI was defined as the tilt of the anterior pelvic plane. The orientation of the pelvis was altered to define a PSI of -30° to 30° at 1° increments. Hip adduction and rotation were fixed in a standing position. Femoral head coverage was quantified in the anterior, superior, posterior, inferior regions for each PSI position.

Results Change in coverage was largest in the anterior region (29.8%) and smallest in the superior region (6.5%). Coverage increased linearly in the anterior region as the PSI increased, while a linear decrease was found in the posterior region and the inferior region (all p<0.001). The slope of the regression line for these regions were 0.513, -0.316, and -0.255, respectively. For the superior region, coverage increased when the PSI was altered from -30° to 5° and decreased when the PSI was larger than 5°.

Conclusions A 1° increase in PSI resulted in an increase of 0.5% in anterior coverage and a decrease of 0.3% in posterior coverage. Our findings provide baseline data that improve our understanding of the effect of PSI on femoral coverage.

ABSTRACT. If appropriate ligament balance is not achieved in total knee arthroplasty (TKA), stiffness or instability of the knee can occur. Active laxity assessment in TKA may be an important consideration for optimal balance during surgery but is rarely performed. Active stress x-rays can be captured preoperatively to define the patient specific coronal laxity. The purpose of this study is to determine if laxity profiles gathered from stress x-rays could be predicted from a patient’s preoperative alignment & bony morphology. 58 patients with an average age of 67 were retrospectively analysed. Preoperative CT scans were captured for each patient to generate 3D bone models from which anatomic landmarks were identified and used to calculate morphological & alignment measurements. Stress x-rays were obtained preoperatively using a Telos (METAX, Germany) device to apply varus & valgus forces at 20o extension and between 45 & 90o flexion. 3D CT-generated bone models were registered to the stress x-rays using Mimics (Materialise, Belgium) and laxity range calculated. Several morphological measures significantly correlated with laxity, the strongest correlation was between laxity midpoint at 20˚ flexion and supine coronal alignment (r = 0.95, p < 0.001). No significant correlations were identified with laxity range at 45-90˚ flexion. Anatomic measurements that correlated significantly with laxity range at 20˚ flexion were coronal angle between the anatomic and mechanical axes of the femur (r = -0.34, p = 0.008), external rotation of Whiteside’s axis from the transepicondylar axis (r = -0.28, p = 0.033) & the supine lateral joint gap (r = -0.31, p = 0.018). Obtaining stressed x-rays to determine functional laxity range is not practical for routine TKA preoperative analysis. Previous investigations have found correlations between knee osteoarthritis, alignment and laxity, but have not investigated 3-dimensional patient specific morphological measures. At this stage knee laxity at 20° flexion has shown a significant correlation, indicating that with greater data, bone morphology may assist in characterising the soft tissue profile. Morphological measurements and anatomic characteristics may help define functional coronal laxity of the knee. Alternate methods in addition to static imaging need to be explored to define patient specific coronal knee laxity. These may include functional knee radiographs or intra-operative measurement techniques or both combined.

ABSTRACT. An important factor which influences the clinical outcome of total knee arthroplasty is the internal/external rotational alignment of the femoral and tibial components. A commonly used reference to align the tibial component rotationally is the tibial tuberosity (TT). An established radiographic measure to quantify the relative rotational alignment of the femur and tibia in extension is the tibial tuberosity-trochlear groove (TT-TG) distance. However, the function of the extensor mechanism in terms of the relationship between TT position, TT-TG distance, and internal/external rotation under weight-bearing conditions is still unknown and accordingly component rotational alignment targets.

The study was based on 8 healthy subjects. Knee kinematics under full-body weight-bearing in single legged stance were available from a combined magnetic resonance imaging approach. In this data, we measured the TT position and TT-TG distance and calculated the internal/external rotation at 4 different knee flexion angles. The relationship between the parameters was investigated by using linear correlation analysis.

The TT-TG distance decreased with greater knee flexion. The linear correlation between initial TT-TG distance and TT position showed a good correlation and indicated that a greater TT-TG distance was associated with an increased lateral TT position. Looking at the kinematics, in 6 of 8 subjects a greater lateral TT position resulted in an increased tibial internal rotation between extension and mid-flexion.

In conclusion, the position of the TT and TT-TG distance are functionally relevant in the weight-bearing kinematics and should be carefully considered in preoperative planning and intraoperatively to avoid any conflicts in joint mechanism.

ABSTRACT. The purpose of this study is to characterize and compare subchondral glenoid bone densities in both non-arthritic (N) and A1, A2, B1, B2 and B3 osteoarthritic glenoids. 150 computerized tomography scans containing equal numbers of glenoids were segmented semi-automatically. Each reconstructed glenoid was divided into four quadrants. Average bone densities were measured at each Volume of interest (VOI), defined at depths of 0 - 2.5 mm (Zone A), 2.5 - 5 mm (Zone B) and 5 - 7.5 mm (Zone C). Osteoarthritic glenoids had higher mean bone densities than N glenoids. Mean bone densities were uniform amongst all quadrants for N glenoids. A2 glenoids had increased bone density measured posteriorly in Zones B and C. In B1 and B2 glenoids, Zones B and C demonstrated increased bone densities of posterior quadrants compared to anterior quadrants. B3 glenoids presented similar results as A1 and A2 glenoids. Cystic changes were more pronounced in anterior quadrants of A2, B1, B2 and B3 glenoids. This study reveals that A1, A2 and B3 glenoids, though geometrically symmetrical, have irregular bony densities similar to B2 glenoids. These findings have clinical implications for reaming the glenoid and implant fixation.

ABSTRACT. Introduction: Although there are many studies on the alignment advantages when using the robotic-arm assisted (RAA) system for total knee arthroplasty (TKA), there have been questions regarding patient-reported outcomes. Therefore, the purpose of this study was to use this index to compare: 1) total; 2) physical function; and 3) pain scores for manual vs. RAA patients.

Methods: We compared 53 consecutive robotic-arm assisted to 53 consecutive manual TKAs. No differences in pre-operative scores were found between the cohorts. Patients were administered a modified WOMAC satisfaction survey pre-operatively, and at 1-year post-operatively. Univariate analyses and multivariate models with stepwise backward linear regression were utilized to evaluate the associations between outcome scores and surgical technique, age, sex, as well as body mass index (BMI).

Results: The RAA cohort had significantly improved mean total (6±6 vs. 9±8 points, p=0.03) and physical function scores (4±4 vs. 6±5 points, p=0.02) when compared to the manual cohort. The mean pain score for the RAA cohort [2±3 points (range, 0 to 14 points)], was also lower than that for the manual cohort [3±4 points (range, 0 to 11 points) (p=0.06)]. On backward linear regression analyses, RAA was found to be significantly associated with more improved total (beta coefficient [β]-0.208, SE [standard error] 1.401, p<0.05), function (β=0.216, SE=0.829, p<0.05), and pain scores (β-0.181, SE=0.623, p=0.063). The RAA technique was found to have the strongest association with improved scores.

Conclusion: This study suggests that RAA patients may have short term improvements at minimum 1-year post-operative. However, longer-term follow up with greater sample sizes are needed to further validate these findings.

ABSTRACT. Background: The purpose of this study was to evaluate hospital admissions for revision surgeries associated with robotic-arm assisted unicompartmental knee arthroplasty (rUKA) vs. manual UKA (mUKA) procedures.

Methods: Patients ≥18 years of age who received either a mUKA or a rUKA procedure were candidates for inclusion and were identified by the presence of appropriate billing codes. Procedures performed between March 1st, 2013 and July 31st, 2015 were used to calculate the rate of surgical revisions occurring within 24-months of the index procedure. Following propensity matching, 246 rUKA and 492 mUKA patients were included. Revision rates and the associated costs were compared between the two cohorts. The Mann-Whitney U test, was used to compare continuous variables, and fisher’s exact tests was used to analyze discrete categorical variables.

Results: At 24-months following the primary UKA procedure, patients who underwent rUKA had fewer revision procedures (0.81% [2/246] vs. 5.28% [26/492]; p=0.002), shorter mean LOS (2.00 vs. 2.33 days; p>0.05), and incurred lower mean costs for the index stay plus revisions ($26,001 vs. $27,915; p>0.05) than mUKA patients. Length of stay at index, and index costs were also lower for rUKA patients (1.77 vs. 2.02 days; p=.0047) and ($25,786 vs. $26,307; p>0.05).

Conclusions: Study results demonstrate that patients who underwent rUKA had fewer revision procedures, shorter LOS, and incurred lower mean costs (although not statistically different) during the index admission and at 24-months post-operative. These results could be important for payers as the prevalence of end-stage knee OA increases alongside the demand for cost-efficient treatments.

ABSTRACT. INTRODUCTION: Unicompartmental knee arthroplasty (UKA) is effective for alleviating symptoms of arthritis in a single knee compartment; however, errors in alignment and instability may predispose to failure. Robotic technology has improved precision, but its impact on functional recovery after UKA remains unknown. The purpose of this study was to compare early functional recovery, pain, and radiographic alignment in UKA performed with either robotic assistance or conventional methods. METHODS: All consecutive patients undergoing UKA by a single physician from January 2015 to March 2018 were retrospectively reviewed. Lower Extremity Functional Scale (LEFS), KOOS Jr, and VAS (0-10) outcomes scores were collected preoperatively, 1, 6, and 12 weeks postoperatively. Radiographic alignment was assessed at the initial postoperative visit. RESULTS: There were 161 patients, 65 conventional and 96 robotic. At baseline, there was no difference in average age, BMI, or outcomes scores between the two groups. Average preoperative pain was significantly higher in conventional patients (6.1 vs. 5.4; p=0.04). At 3 weeks post-op, conventional UKA patients still had significantly higher pain levels (3.9 vs. 3.1; p=.02). Both groups showed significant improvement in LEF (p<.0001) scores over time (p<.0001). Significant improvement in KOOS Jr. scores from pre-op (52.3) to 6 and 12-weeks postoperatively (67.6, 69.8; p<0.001). When comparing PCS of the VR/SF-12, the robotic-assisted group had significantly higher improvement compared to the conventional group at 6-weeks (40.6 vs. 35.6; p=0.02). CONCLUSION: Robotic assisted UKA resulted in fewer radiographic outliers, and more rapid recovery with less early postoperative pain although functional differences tended to equilibrate by 3 months postoperatively.

ABSTRACT. Total knee arthroplasty is a successful procedure. However, there is still area for improvement as up to 15-20% of patients remain unsatisfied. Robotic-assisted surgery (RAS) may improve patient outcomes by providing a reproducible way of obtaining neutral mechanical alignment of the limb, which has been shown to reduce early revisions and correlate with patient reported outcomes after surgery. We prospectively enrolled 106 patients undergoing robotic-assisted TKA by a single surgeon performing a measured-resection femur-first technique using the OMNIBotic system. Patients completed a KOOS and New Knee Society Score (KSS) pre-operatively and at 3, 6, 12, and 24 months (M) postoperatively. Changes in the five KOOS sub-scales were compared to available literature data from the FORCE – TJR cohort, as well as to individual studies reporting on conventional and computer-assisted TKA. When compared to FORCE-TJR 6-month (M) and 2-year (Y) data, the RAS cohort had significantly higher improvements at 6M for pain (40.5 vs. 31.1, p<.001) and at 2Y for all five KOOS sub-scores. The larger improvement was due to the RAS cohort having lower baseline KOOS scores than the FORCE-TJR cohort, except for the Sports-Recreation subscore, which was similar pre-operatively but significantly higher post-operatively for the robotic cohort. Despite having poorer joint function and higher pain pre-operatively, robotic-assisted TKA patients achieved excellent self-reported outcomes, with significantly higher levels of improvement through two years post-surgery when compared with large national cohort studies. Further controlled clinical studies are warranted to determine if these results translate to other groups of surgeons, centers and patients.

ABSTRACT. Haptic robotic-arm assisted technology improves accuracy in unicompartmental knee replacement through utilizing a preoperative 3-D plan, optical navigation for real-time intraoperative feedback on soft tissue laxity, and robotic arm for precise bone preparation. This technology became clinically available for total knee arthroplasty (TKA) in 2016. We present outcomes from the early adoption of this technique. A retrospective chart review compared data from the first 120 robotic-arm assisted TKAs performed December 2016 through July 2018 to the last 120 manually instrumented TKAs performed May 2015 to December 2016, prior to robotic technology adoption. Robotic surgery was associated with significantly increased anesthesia (212 vs 187 mins, p < 0.01) and operative (135 vs 112 minutes, p < 0.01) time. The robotic group had a lower hospital length-of-stay (2.7 vs. 3.4 days, p < 0.001). Discharge to home was not statistically different between robotic and manual groups (89% vs. 83%, p = 0.2). Robotic technology was associated with decreased variability in implant positioning, with smaller variances in the lateral distal femoral angle (LDFA; 3.5 vs 6.6 degrees, p < 0.01) and posterior tibial slope (1.8 vs. 5.3 degrees, p < 0.01). Mean limb alignment, as measured by tibiofemoral angle, was slightly less valgus in the robotic group (3.9 vs 4.4 degrees, p = 0.09). Postoperative range of motion was significantly increased for robotic-arm assisted TKA patients, with less flexion contracture at 2-weeks (1.8 vs. 3.3 degrees, p < 0.01), 7-weeks (1.0 vs. 1.8 degrees, p < 0.01), and 3-months (0.6 vs 2.1 degrees, p = 0.02) post-surgery. Postoperative Knee Society scores were similar between groups. Preliminary findings demonstrate robotic-arm assisted TKA is safe and efficacious with outcomes comparable, if not superior, to that of manually instrumented TKA.

ABSTRACT. Automated image segmentation of surgical tissue could provide new control inputs into Computer Aided Orthopaedic Surgery (CAOS) systems, leading to more intelligent and adaptive surgical devices. This work proposes a machine learning based system that classifies tissues based on the surface response to excitation with a laser pointer. Training data is acquired from cadaveric tissue of skin, bone, muscle and fat being intersected with a 650nm laser, with images captured by a 1MP camera. Isolating the pixels in each image corresponding to the laser pointer is done using greyscale brightness thresholding. The images are subsequently passed through a Convolution Neural Network (CNN) for training. Validation of the network is done with 2000 unseen images to quantify the accuracy. Using googlenet’s CNN architecture the overall accuracy is 96.9%, with an accuracy of 97.8%, 99.2%, 93.4% and 97.0% for bone, skin, fat and muscle tissues, respectively. Misclassification occurred due to the resizing procedure altering the patterns within the images, particularly affecting the fat tissue. The radius of the laser scatter pattern is also indicative of the tissue, but this information is lost during the resizing procedure. The overall results are very good, and with further research and optimisation, a new technique to automatically classify tissue types intraoperatively could be implemented.

ABSTRACT. The integration of augmented-reality (AR) in medical robotics has been shown to reduce cognitive burden and improve information management in the typically cluttered environment of computer-assisted surgery. A key benefit of such systems is the ability to generate a composite view of medical-informatics and the real environment, streamlining the pathway for delivering patient-specific data. Consequently, AR was integrated within an orthopaedic setting by designing a system that captured and replicated the user-interface of a commercially available surgical robot onto a commercial head mounted see through display. Thus, a clinician could simultaneously view the operating-site and real-time informatics when carrying out an assisted patellofemoral-arthroplasty (PFA). The system was tested with 10 surgeons to examine its usability and impact on procedure-completion times when conducting simulated PFA on sawbone models. A statistically insignificant mean increase in procedure completion-time (+23.7s, p=0.240) was found, and the results of a post-operative qualitative-evaluation indicated a strongly positive consensus on the system, with a large majority of subjects agreeing the system provided value to the procedure without incurring noticeable physical discomfort. Overall, this study provides an encouraging insight into the high levels of engagement AR has with a clinical audience as well as its ability to enhance future generations of medical robotics.

ABSTRACT. Registration is one of the key steps in computer assisted orthopaedic surgery as it defines the location of the patient in the surgical system so that subsequent assistance or navigation can be performed correctly. Currently, the mainstream of orthopaedic registration is based on optical markers in order to achieve registration and real-time tracking together. However, the use of optical markers may also cause issues such as increased operating time or bone complications due to the invasiveness of fixation pins.

The development of depth sensing technology makes it possible to capture spatial geometry in real time, which opens the possibility of markerless registration for orthopaedic surgery using affordable depth cameras. As depth cameras capture the geometry of the environment as a whole, the first step to make the depth images usable for registration is to extract the part that belongs to the target bone from the environment. To this end, in this paper we developed a depth image segmentation method based on deep learning. We chose knee surgery as the study object and created our own labelled dataset from a cadaveric knee. The segmentation network adopts the “U-Net” architecture that can perform end-to-end segmentation without pre-processing. After training, a pixel-wise segmentation accuracy of 91% is achieved on the testing dataset, which could provide sufficiently accurate segmented points for markerless registration in the future.

ABSTRACT. With the increasing spread of computer assisted surgery, more and more modern operating rooms are equipped with navigation systems, each coming with its own tracking camera. Since those cameras are part of the closed monolithic navigation system, they can’t be used for other applications than the one intended by the supplier. With the novel service oriented device connectivity standard (IEEE 11073-SDC), introduced by the OR.NET initiative (www.ornet.org), needless double procurements could be avoided and multiple systems could use the same camera that – similar to OR lights – could be installed as a standard equipment in each OR. This would decrease the cost-to-benefit ratio also of new applications that would currently as such not justify to acquire a proprietary tracking camera. While the integration of a tracking camera to an open medical device IT network can open up for new applications, it should on the other hand not impair the usability and the safety of the navigation system. Therefore, a low latency must be guaranteed between tracking camera and navigational display. This paper evaluates the integration of an atracsys fusionTrack 500 tracking camera into the OR.NETwork. The response time from a change in the real world to the reception of the corresponding data package is measured to determine the feasibility of an integration without impairing current navigational tasks. The results show that, as long as the underlying network infrastructure is not at its capacity limit, latencies below 60 ms are achieved. Therefore, the integration of a tracking camera for navigational tasks is feasible.

ABSTRACT. We report on the technical assessment and first clinical use of a new 3D capable C-arm equipped with a flat-panel detector – the Cios Spin (Siemens Healthineers). Imaging performance was characterized through objective metrics of image quality and dose, and clinical use tests provided assessment in a range of common orthopaedic interventions. The system contained four imaging protocols for acquiring 3D images. Dosimetry was assessed at each imaging protocol using 16 cm and 32 cm acrylic CTDI phantoms to simulate an adult head and body, respectively. The absorbed dose (air kerma, Dw) was calculated as the weighted sum of air kerma measured at central and four peripheral locations in the CTDI phantoms. Image quality was assessed in terms of spatial resolution (modulation transfer function, MTF), image noise (noise-power spectrum, NPS), and contrast-to-noise ratio (CNR). An IRB approved clinical study (17 patients) was conducted to evaluate the usability of the system for 2D and 3D imaging in orthopaedic interventions. Interventional tasks were evaluated by a fellowship-trained orthopaedic surgeon in terms of workflow, image quality, and ease of use. The dose (Dw) varied according to imaging protocol over a range of ~0.9-5.4 mGy and ~7.4-43 mGy for adult head and body protocols, respectively. The MTF was consistent with sub-mm spatial resolution, and noise characteristics were consistent with bone visualization. Initial clinical studies demonstrated the potential benefits of the C-arm for planning and quality assurance / verification as well as potential improvements to workflow and dose reduction to surgical staff. The studies provide a quantitative characterization of imaging performance and dose, providing a valuable guide to technique selection and clinical deployment.

ABSTRACT. Confidence-weighted structured phase symmetry (CSPS) is a state-of-the-art bone segmentation technique for ultrasound (US), which has been recently proposed for automatic diagnosis for developmental dysplasia of the hip (DDH). However, CSPS relies on complex image phase feature analysis which is computationally expensive, and in our preliminary tests we have found it to be sometimes inaccurate. We evaluate a simpler alternative segmentation technique which we previously published, called Shadow Peak (SP), which uses intensity analysis to perform fast and accurate US bone segmentation. On average, SP segmentation ran 15 times faster for 2D US images, when tested on 15 hip images of pediatric patients. Furthermore, SP improves the segmentation F-score to 94%, compared to 72% when using CSPS segmentation.

ABSTRACT. For many years, surgical navigation systems have been used in orthopaedics. These solutions however often mean additional time and complexity because, essentially, of the markers. We want to introduce a new solution based on depth camera which could be used intraoperatively to estimate the 3D pose of surgical instruments without specific markers. The goal of this paper is to assess, on synthetic data, an algorithm called Clustered Viewpoint Feature Histogram (CVFH) to estimate the pose of an orthopaedic cut guide used during knee surgeries. A specific simulator has been developed for this study which allows the simulation of a point cloud associated to the cut guide. The Average Distance Distinguishable (ADD) metric has been measured 1000 times according to several cut guide orientations and several noise levels. The success rate has also been analyzed. It is commonly considered that the pose is correctly estimated if ADD is less than 10% of the largest dimension of the object. The ADD metric and the success rate vary from 2.12 ± 4.46 mm to 2.82 ± 5.73mm and from 96.0% to 92.4% for respectively a low (0 mm) and a high noise (10mm). The results are very encouraging. However, more parameters have to be assessed. Similarly, the accuracy and reliability of such method have to be evaluated in a real clinical environment.

ABSTRACT. Early diagnosis of sacroiliitis may lead to preventive treatment which can significantly improve the patient's quality of life in the long run. Oftentimes, a CT scan of the lower back is acquired for suspected back pain. However, since the differences between a healthy and an inflamed sacroiliac joint in the early stages are subtle, the condition may be missed. We have developed a new automatic algorithm for the diagnosis and grading of sacroiliitis CT scans as incidental findings that is based on supervised machine and deep learning techniques. The input is a CT scan that includes the patient's pelvis. The output is a diagnosis for each sacroiliac joint. Experimental results on 242 cases yield a binary and a 3-class case classification accuracy of 92% and 81%, a sensitivity of 95% and 82%, and an Area-Under-the-Curve of 0.97 and 0.57, respectively. Automatic computer-based analysis of CT scans has the potential of being a useful method for the diagnosis and grading of sacroiliitis as an incidental finding.