An Effective Multiclass Human Skin Lesion Diagnosis System Based on Convolutional Neural Networks

Recently, deep learning algorithms have acquired considerable attention to diagnosing different human diseases. Hence, recent researches prove the efficiency of these algorithms in skin lesions diagnosis using dermoscopic images. However, the situation of multiclass skin lesions is not taken into consideration via most of such researches. In this paper, an effective system of multiclass human skin lesion diagnosis based on convolutional neural networks (CNNs) is proposed. This proposed system is designed with multilayers, implemented, and calibrated for classifying the images of skin lesions into seven categories: basal cell carcinoma, actinic keratoses, dermatofibroma, benign keratosis, vascular, melanocytic nevi, and melanoma skin lesions. The proposed CNN based diagnosis system is evaluated via the experiments on the HAM10000 dataset using different terms. The obtained results illustrate that the proposed diagnosis system exceeds most of the recent existing systems, depending on the chosen terms involving precision (84%), recall (82%), F1-score (81%), and accuracy (95%).