A Transparent Search-Based Framework for Skin and Breast Cancer Diagnosis Using Siamese Networks

The detection and classification of cancerous lesions, including skin and breast cancer, have become critical tasks in medical image analysis. This research aims to develop an automated system that classifies breast histopathology images and skin lesion images using advanced Convolutional Neural Networks (CNN), enhanced with Residual Blocks and Convolutional Block Attention Module (CBAM). The dataset comprises images of benign and malignant tumors for breast cancer and seven types of skin lesions. The system utilizes CNN-based models trained with enhanced image augmentation techniques to handle class imbalances and improve performance. The model’s architecture also integrates Siamese Networks combined with Support Vector Machines (SVM) for better discrimination between classes. The results demonstrate high accuracy in detecting malignant tumors and classifying skin lesions. This system promises to assist healthcare professionals by offering a reliable and efficient tool for automated cancer detection, reducing human errors and diagnostic time. The study also explores focal loss, class weights, and oversampling techniques to deal with data imbalance. The model's performance was evaluated using confusion matrix metrics, F1-score, and accuracy, showing promising results in distinguishing between benign and malignant cases.

Keywords: CNN, SVM, Siamese Networks, Breast Cancer, Skin Lesion, Malignant, Benign, Image Augmentation, Residual Blocks, CBAM