LesionNet-FCL: Deep Learning Framework for Detecting and Classifying Skin Lesion

Abstract

Early detection is important for skin cancer as it can increase both the life expectancy of patients and the likelihood that their acquired treatments will work. This study presents a novel deep-learning-based framework called LesionNet- FCL, which was created to categorise types of skin lesions. The work utilises EfficientNet-B0 as a feature extractor by extracting discriminative features from dermoscopic images passed as inputs to a classifier that is a Fully Connected Layer (FCL). Regarding the evaluations, the model will be tested using the ISIC data set with all data obtained from Kaggle. To further support the evaluation processes, the study also included many pretrained CNN's with the same experimentation conditions, ultimately resulting in LesionNet- FCL outperforming all of the other models by generating an overall classification accuracy of 96%. The overall results indicate that combining the process's optimized classification method with the use of deep features will generate superior results than an individual model. Therefore, supporting the need for dermatologists to have reliable computer-aided diagnostic assistance as a means to improve still-image-based clinical decision-making and to enable faster identifications.