Skin Disease Classification Using Deep Learning

Abstract

Timely detection of skin disease becomes vital in avoiding serious medical consequences. Unfortunately, there is a shortage of dermatology experts, and therefore an automated system is needed. In this research, an efficient deep learning architecture called EfficientNetB0 is considered to perform the classification of seven skin lesions types utilizing the HAM10000 dataset. As medical data usually contains noise and class imbalance, certain pre-processing methods have been applied, like removing artefacts and normalizing the data. In order to balance the distribution of classes for better training of a model, the authors applied an altered loss function. To evaluate the efficiency of the proposed approach, not only accuracy but also other metrics were taken into account. The findings of this study prove the ability of the system to distinguish between skin diseases.