Skin Cancer Classification Using Premature Convergence Strategy-based Artificial Jelly Search Optimization with Convolutional Neural Network

Melanoma, the deadliest type of skin cancer in humans, poses significant challenges while diagnosing skin lesions through thermoscopic imaging.These challenges include issues such as hair interference, inaccurate capturing of skin lesion shapes, and extraction of irrelevant features.However, overfitting occurs due to oversampling, resulting in erroneous skin cancer captures, and failure to detect noise and unwanted hair in melanoma images.The research proposes the Premature Convergence Strategy with Artificial Jelly Search Optimization (PCSAJSO) algorithm and Convolutional Neural Network (CNN) for classifying skin cancer into melanoma and non-melanoma.Initially, images are obtained from ISIC2018, ISIC2019, and PH2 datasets, used to analyze the performance of the proposed method.The W-Net method is utilized to segment skin cancer by capturing the contextual information from various scales in skin cancer images.The PCS technique ensures a balance between exploration and exploitation in search space which helps to increase convergence speed and adjusts population size for skin cancer classification.In comparison to the existing techniques like Inception-ResNet and U-Net, the PCSAJSO-CNN achieves commendable accuracies of 98.15%, 99.38%, 99.70%, 94.68%, and 97.08% respectively on the ISIC2016, ISIC2017, ISIC2018, ISIC2019 and PH2 datasets.