DermaVision-CNN: An Intelligent Deep Learning Framework for Skin Cancer Detection using Dermoscopic Images

Abstract

Skin cancer is a growing and a health concern worldwide. The key is early detection for the patient. This paper proposes an automated skin cancer identification and grouping system called Derma Vision-CNN, using dermoscopic images with the assistance of a smart deep-learning system. The method combines image cleaning, image augmentation, deep feature extraction, and final sorting, ultimately improving the diagnostic effectiveness. On the cleaning side, images are resized, normalized, filtered to reduce noise and hair lines in the image with morphological black-hat operations and then contrast enhanced. The model is then prevented from overfitting and augmented with variability via rotating, flipping, scaling, and zooming the dataset. The cleaned images then go into a series of convolutional, pooling and fully connected layers, which learn features and label images independently. The final decision, between a skin lesion and a non-cancerous case, is made with the use of a SoftMax-based output layer. The experimental results demonstrate the excellent classification performance of Derma Vision-CNN with an accuracy of 98.4%, surpassing the results obtained by traditional classifiers and other deep models. Combined, it proves to be a intelligent, scalable and trusted skin cancer diagnostic solution.