Deep CNN techniques have dramatically become the state of the artin image classification. However, applying high-capacity Deep CNN in medicalimage analysis has been impeded because of scarcity of labeled data. This studyhas two primary contributions: first, we propose a classification model toimprove performance of classification of skin lesion using Deep CNN and DataAugmentation. Second, we demonstrate the use of image data augmentation forovercoming the problem of data limitation and examine the influence of differentnumber of augmented samples on the performance of different classifiers. Theproposed classification system is evaluated using the largest public skin lesiontesting dataset, containing 600 testing images, and 6,162 training images. Newstate-of-the-art performance result is archived with AUC (89.2% vs. 87.4%), AP(73.9% vs. 71.5%), and ACC (89.0% vs. 87.2%). In additional, we explore theinfluence of each image augmentation on the three classifiers and observe thatperformance of each classifier is influenced differently by each augmentationand has better results comparing with traditional methods. Thus, it is suggestedthat the performance of skin cancer classification and medial image classifica-tion could be improved further by applying data augmentation.

Research paper: Deep CNN And Data Augmentation For Skin Lesion Classification

Tri Cong Pham, Chi Mai Luong, Muriel Visani, Van Dung Hoang