A grad-CAM and deep learning based image classification for skin diseases

Abstract

The classification as well as identification of skin diseases are crucial focal points in deep learning research. This initiative aims to enhance diagnostic systems through utilising deep convolutional neural networks (CNNs) and employing sophisticated architectures such as ResNet-50 and VGG-16 to achieve higher accuracy. We utilise Grad-CAM to enhance the visualisation of our model's predictions. Our technique utilises the ISIC 2019-2020 dataset, which consists of 9,200 photos. It entails thorough training and testing. In addition, Our experimental approach is improved by the ISIC Archive dataset, allowing us to conduct detailed performance evaluations. The accuracy of our customised Convolutional Neural Network (CNN) model on the ISIC 2019-2020 dataset was 87%. In comparison, the CNN model combined with ResNet achieved an accuracy of 95%, while the CNN model combined with VGG16 produced a precision of 92%. CNN+ResNet and CNN+VGG16 models were developed using the ISIC Archive dataset as their basis, yielded accuracy values of 92% and 94%, respectively while the proprietary CNN model attained an accuracy of 91.91%. The results of this study provide a scalable and efficient framework that surpasses current approaches, enhancing the deep learning-based dermatology diagnostic systems' interpretability and accuracy.