Unified Approaches For Skin Cancer Detection: A Systematic Aspect Of Review Of Support Vector Machines To Resnet-50, And Deep Learning Models

Abstract

It is still a big challenge for doctors to find skin cancer early and especially melanoma. Over the past few years, artificial intelligence (AI) has begun to greatly improve dermatological diagnostics and support better clinical decisions. We carried out a review that evaluates traditional Support Vector Machines (SVMs) against deep learning methods, like RESNET-50, when dealing with dermoscopic images. A review of peer-reviewed articles shows that deep learning models achieve greater performance than traditional ones, and RESNET-50, in particular, beats the average with accuracy greater than 92%, sensitivity up to 94.3%, while its AUC value is above 0.96 in some large-scale datasets. Still, both SVM and hybrid systems like SVM with texture and color show high discriminative power and efficiency when dealing with simpler and low-resource datasets. Improvements in performance have not solved all the problems. Because clinical use often requires understanding a model, and annotating many datasets is costly for healthcare, deep learning models have a limited use in hospitals. Not having enough evidence in the training process, especially for darker skin, results in this bias during diagnosis. Even though these new approaches look promising, they are both limited by the availability of compatible hardware and by vague regulations. This study points out that we need AI systems that are strong, clear in what they do, and use the same standards, with attention to ensuring fairness and transparency in their decisions. Other possible future steps discussed in the paper are explainable AI, federated learning, and cross-domain model generalization, designed to take skin cancer findings from experiments into real medical practice.