Kolmogorov-Arnold Networks in Dermatology: Skin Lesion Classification

The increased levels of pollution have paved the way for increased global warming. The increased levels of temperature have given a way to different diseases and severe conditions which need immediate medical assistance. One such organ is our skin as it is in constant contact with the environment and gets the maximum exposure also. Skin diseases are usually not considered much severe because they do not cause any inflammation (as an indication) but they can be cancerous also. The main problem is the accurate identification of particular skin diseases. This becomes a problem because of people's different complexion. This study employs the Kolmogorov-Arnold Network (KAN) to classify these lesions as benign and malignant so that further investigation can be carried out. KAN is considered to be an alternative to multilayer perceptron(MLP). This study also compares the accuracy of the above two algorithms for skin lesion classification in which KAN outperforms MLP with an accuracy of 89.5%.