Accurate melanoma subtype classification via deep learning and dermoscopic images

Acral lentiginous melanomas Skin cancers such as melanoma are particularly dangerous. The likelihood of melanoma being totally cured increases with time from diagnosis. The importance of patient and clinician understanding of early melanoma symptoms cannot be overstated. Dermatologists can diagnose it with dermoscopic images. Comparison of precision, recall, f - measure, and accuracy between Stages 1 and 2 of melanoma and benign nevi because of the small differences between the two. Skin cancer studies tend to focus on determining if a lesion is melanoma or non-melanoma. Moreover, only a small amount of research has been done on the classification of subtypes of melanoma. In this study, deep learning and dermoscopic imaging were employed to identify melanoma subtypes like AM. In this research, we presented a new deep-learning model for organizing skin cancer. We categorized skin lesions from the "Yonsei University Healthcare System in South Korea" using a dermoscopic image collection. A robust automatic system for detecting acral melanoma was built through data augmentation. The evaluation of our model's performance was based on transfer learning. This study employs three pre-trained models: VGG16, the Inception V3 network, and Dense Net, which are then assembled via stacking. By using stacking, VGG 16 achieved an accuracy of 91.7241, Inception 3 achieved an accuracy of 93.1, Dense Net achieved an accuracy of 94.48, and Ensemble these models by using stacking achieved an accuracy of 97.2%.