Publication

Abstract : The key to increasing the survival rates of patients with melanoma is the correct diagnosis of the disease at the initial stages. This article presents a detection system, which is automated, based on deep learning algorithms used over dermoscopic images. These models are advanced architectures to suggest a comparative framework that we are going to utilize: a custom CNN, VGG16, ResNet50, MobileNetV2, and InceptionV3. The models were trained on the ISIC dataset with the addition of optimized preprocessing, data augmentation and strategic fine-tuning. The best performer with high accuracy of 92 percent and 89.5 percent were VGG16 and ResNet50 respectively. The system that has been developed shows much promise in terms of accurate melanoma detection in clinical settings.