A Robust Shallow CNN Architecture for Performing Ablation Studies on Skin Cancer Images

Abstract

Skin disease is an increasingly common form of disease that affects millions of people worldwide every year. It is caused by the uncontrolled growth of abnormal skin cells. Skin disease detection is an important area of medical research as skin diseases can have a major impact on the quality of life of patients. As a result of a significant amount of data available for model training and improved model designs, Deep Learning techniques have grown rapidly for computer vision applications. This study aims to describe a robust deep-learning CNN model that categorizes skin disease using into six classes based on a deep learningbased CNN. The uninvited regions of skin disease are removed, the image is enhanced, and the disease is tinted by removing artefacts, reducing noise, and improving the image. The augmentation techniques have increased the number of skin disease images. Initially a base CNN model has been proposed in the augmented dataset. An ablation study has been employed to get the robust CNN model, which name is SkinNet-11. The model is trained with a set of publicly available skin disease images. The proposed robust SkinNet-11 achieved the best results with 98.00% accuracy. The model is robust and shows a high degree of generalizability on unseen data. The model also achieves a high level of precision and recall in both binary and multi-class skin disease detection scenarios