Deep learning algorithm based on YOLOV5 Neural Network for dermatoscospic classification and detection of epithelial cancer (MELANOMA)

Resumo

Technical concepts are introduced about epithelial cancer or melanoma, one of the most common and
aggressive types of existing cancers, and about the use of applications or tools based on Artificial Intelligence
techniques and the influence of the application of machine learning to perform pre-diagnoses in medicine in
general, which accelerate the time of discovery of the disease and exponentially increase the chances of cure of
the affected patient. The technical details of the algorithm based on Deep Learning Yolov5 are developed. The
possibility of applying this tool for various purposes is discussed. The dataset used in this research is analyzed and
the possibilities of characterizing tumors between benign and malignant are verified, as well as the types of
epithelial formations found are classified. The possibility of using these classifications as input for algorithms
based on deep learning by scanning images is discussed. It is concluded that it is possible to create logical filters
that use the readings of the output data provided by the execution of the Yolov5 algorithm on a photographic basis
of dermoscopic exams, obtaining relevantly high levels of accuracy that would point out with great precision the
possible existence of cancerous formations, thus facilitating the pre-diagnosis and early combat of skin cancer,
considerably increasing the survival of patients.