Deep Neural Nets in Real-Time Detection of Cancerous Epidermal Lesions

Skin cancer is among the most common malignancies amid humans. There are three basic types of skin cancers, i.e., basal-cell carcinoma, squamous cell carcinoma, and melanoma. Among the three basic types, melanoma has the lowest surviving rate however, basal-cell carcinoma is the most occurring cancer subtype. Therefore, early detection of skin cancer is necessary to increase the chance of survival of cancer patients. Since it has more prominent visual symptoms than other diseases, computer vision could be used to detect the advent of cancer early on. This study uses the YOLOv4 object detection algorithm to classify skin lesions into one of the three classes - basal cell carcinoma, squamous cell carcinoma, and melanoma. The training images are first manually annotated and then fed to the algorithm for object detection and classification. The results show that the YOLOv4 algorithm is not only accurate in detecting the skin cancer subtype but can also be integrated with dermoscopy IOT devices for real-time cancer detection by dermatologists.