Neural Network in Dermatology: Pioneering Skin Lesion

Title:Neural Network in Dermatology: Pioneering Skin Lesion

Authors:Nasif Fuad, Abdur Rakib, Abdullah Al Noman, Md Rafiqul Islam, Mahbubur Rahman and Md Abu Rumman Refat

Conference:STI 2024

Tags:Augmentation, CNN, computer vision and pattern recognition, confusion matrix, convolutional neural network, convolutional neural networks, data augmentation, data augmentation techniques, deep convolutional neural network, deep learning based, deep learning based method, deep learning models, deep learning techniques, deep transfer learning, detection and classification, detection of skin cancer, Ensemble, ensemble cnn model, ensemble cnn models, ensemble model, global average pooling layer, HAM10000, learning and deep learning, machine learning and deep, machine learning techniques, model s performance, optimal weight combination, pigmented skin lesions, skin cancer, skin lesion segmentation and classification, skin lesions and SMOTE

Abstract:

Skin cancer is one of the most common cancer types in the world. Early detection of skin cancer is crucial as it significantly increases the possibility of successful treatment. Yet, recognition of skin lesions can be challenging because there is a high degree of similarity in the lesion’s appearance, location, color, and size. In response to this challenge, we developed a deep learning based method for skin lesion classification by ensembling two pre-trained CNN models called ResNet50 and DenseNet201 using the HAM10000 dataset. With the imbalance in the dataset, we used the Synthetic Minority Over-sampling Technique(SMOTE) and class weighted approach along with the data Augmentation approach. As a result, our study demonstrates that the proposed approach attained a remarkable 96.0% accuracy on the HAM10000 dataset, surpassing other cutting-edge methods.