Multi-Class Primary Morphology Lesions Classification Using Deep Convolutional Neural Network

Skin diseases are becoming the most prevalent health concern among all nations worldwide. Recognition of skin lesion, abnormal change usually caused by disease or infection in the skin is the first step in diagnosing skin diseases. In dermatology, morphology skin lesions occur due to the disease process's direct result and indicate categorizing a skin lesions' structure and appearance. In this work, we focus on primary skin lesion classification, particularly early-stage detection, and present a deep learning approach to classify images containing skin lesions, macule, nodule, papule, plaque pustule, wheal, and bulla. We applied deep learning techniques for classifying such images into seven classes covering the aforementioned types of lesion. In particular, we performed experiments on pre-trained deep convolutional neural network models to find the most accuracy one. The result shows that the pre-trained model ResNet-50 after the training and testing can achieve satisfactory accuracy of 85.95%.