Proposed Visual Explainable model in Melanoma Detection and Risk Prediction using Modified ResNet50

This study proposed an enhanced visual explainable model for melanoma detection and risk prediction. We utilized the HAM10000 dataset, applying pre-processing techniques to improve image quality. Feature extraction and segmentation were performed using a U-Net model-based Dual Stream CNN-Transformer technique. Feature selection was optimized using the Henry Gas Solubility Optimization (HGSO) algorithm and the Water Strider Algorithm (WSA). A Deep Learning Model (DLM), specifically the Optimal Multi-Attention Fusion (MAF) ConvNeXt, was trained for melanoma detection. For disease severity prediction, we introduced a Modified ResNet-50 model combined with the Explainable AI technique Grad-CAM, providing visual explanations for the model's predictions. Experimental results demonstrate a robust classification performance with an AUC of 0.997, recall of 99%, and precision of 99.5%. This study aims to diagnose an accurate, efficient, melanoma and risk assessment. The Algorithm source code can be accessed at https://github.com/SarvachanVerma/Visual-Explanible-code-for-Melanoma_Matlab