DeepSkin: A Deep Learning Approach for Skin Cancer Classification and Detection

Abstract

One of the diseases that propagate fastest across the globe due to lack of adequate health facilities is skin cancer. Successful prevention is only possible through early and proper diagnosis; however, dermatologists often face difficulties with early diagnosis. DL has simplified the process of identifying skin cancer. CNNs are the most successful at object location and classification. In this study, the HAM10000 dataset that consists of 10,015 samples of seven types of skin lesions is employed. Preparation methods such as sampling, DullRazor to remove hair, and segmentation using an autoencoderdecoder approach enhance the quality of pictures. A number of YOLO models including YOLOv5, YOLOv6, YOLOv7, and YOLOv8 are employed to detect lesions. Some of the DL architectures that are used in categorization include ResNet150, DenseNet169, VGG16, DenseNet201 and Xception. Among them, Xception was the most accurate, indicating that it was superior in feature extraction. This research enhances the precision of diagnosing skin cancer by employing advanced detection and classification algorithms. It assists in early intervention and improved patient outcomes