Related Weighted Feature Subset Model for Skin Cancer Classification using Resnet50 Model

Abstract

Among all diseases, skin cancer is among the deadliest disease if not detected early as per the records from World Health Organization (WHO). Early and accurate skin lesion classification could improve clinical decision-making by facilitating more precise disease identification, which in turn could improve treatment outcomes by halting the progression of cancer. The bulk of skin disease images utilized for training purposes are imbalanced and in low supply, which makes automatic skin cancer classification challenging. Additionally, the model's cross-domain adaptability and robustness pose key obstacles. In order to address these concerns and obtain satisfying results, numerous deep learning-based approaches have recently been extensively utilized in skin cancer categorization. However, there is a dearth of reviews that address the aforementioned cutting-edge issues with skin cancer classification. Using features taken from pre processed images in the publicly accessible datasets, this research intends to build a deep learning based model that can accurately classify skin cancer. It is well-known that pre processed features capture visual qualities pertinent to the classification task, making them more informative than raw image data. Improving the accuracy and interpretability of deep learning classification, this research proposed a model that detects the diseases more accurately from dermoscopy images. The 50-layer ResNet is constructed using a bottleneck architecture. With ResNet50, users can train a state-of-the-art image classification model on massive datasets. The utilization of residual connections is a major innovation that enables the network to learn a set of functions that translate the input into the intended output. By utilizing these residual connections, the network is able to overcome the issue of vanishing gradients and learn more complex topologies than before. This research proposes a Related Feature Subset Model using ResNet50 for Skin Cancer Classification (RFSM-ResNet50-SCC). The proposed model when compared with the traditional models performs better in feature subset generation and skin cancer classification.