A Comprehensive Evaluation and Benchmarking of Convolutional Neural Networks for Melanoma Diagnosis

Melanoma is the most invasive skin cancer with highest risk of death. While it a serious cancer, highly curable if detected early. diagnosis difficult, even for experienced dermatologists, due to wide range morphologies in lesions. Given rapid development deep learning algorithms melanoma diagnosis, crucial validate and benchmark these models, which main challenge this work. This research presents new benchmarking selection approach based on multi-criteria analysis method (MCDM), integrates entropy preference ranking organization enrichment evaluations (PROMETHEE) methods. The experimental study carried out four phases. Firstly, 19 convolution neural networks (CNNs) are trained evaluated public dataset 991 dermoscopic images. Secondly, obtain decision matrix, 10 criteria, including accuracy, classification error, precision, sensitivity, specificity, F1-score, false-positive rate, false-negative Matthews correlation coefficient (MCC), number parameters established. Third, PROMETHEE methods integrated determine weights criteria rank models. Fourth, proposed framework validated using VIKOR method. obtained results reveal that ResNet101 model selected as optimal our case data. Thus, presented proven be useful at exposing targeting ease process proper convolutional network architecture.