Development of Image Classification System to Detect Skin Lesion using Deep Learning Techniques

Skin lesions are abnormal growths or appearances of the skin and may be an indication of several diseases, both benign and malignant. To achieve the correct diagnosis and proper treatment of these lesions, need to be discovered early and diagnosed accurately. Conventional methods of diagnosis, including biopsies and examinations of the dermatologist's eye, are often invasive and expensive, thus limiting access to them in underdeveloped regions. This work proposes a deep learning-based image classification system for the identification of skin lesions. The categorization of dermoscopic images was done using CNN architectures such as ResNet50, DenseNet201, VGG 19 and Xception. ResNet50 performed better than the others in identifying and categorizing skin lesions, achieving the highest accuracy of 87.72%. This non-invasive method and accessibility could greatly improve the diagnostic capabilities while reducing reliance on conventional techniques.