Aplicação Mobile para classificação do Carcinoma Basocelular e Melanoma utilizando Deep Learning

Abstract

This work presents the development of a Deep Learning model using the MobileNetV2 architecture, based on Transfer Learning, for the classification of skin lesions, especially Basal Cell Carcinoma and Melanoma. The problem addressed is the difficulty in the early diagnosis of skin cancer, especially in regions with limited access to specialists. The main objective is to obtain/train an efficient model for the automatic classification of skin lesions and implement it in a mobile application, to assist doctors and patients in the diagnosis and monitoring of skin lesions. For this, data from the ISIC 2020 challenge were used, consisting of 7800 images divided between the Basal Cell Carcinoma and Melanoma classes. The methodological steps included preprocessing of the images, with resizing and normalization, and application of the Cross-Validation technique to validate the model. The model was implemented and trained, achieving an average accuracy of 93.67% across the training subsets and an accuracy of 94.24% with test data. A mobile application was developed to capture images of lesions and classify them based on the trained model as a way to support early diagnosis of skin cancer