Enhancing Skin Cancer Classification on the PH2 Dataset Through Transfer Learning Technique

Abstract

Skin, the largest organ of the human body, serves as a crucial barrier against external threats. Among the myriad skin diseases, melanoma, or skin cancer, stands out as one of the most perilous and lethal conditions. However, its prognosis dramatically improves when detected early. The advent of advanced diagnostic imaging methods has mitigated the risks associated with cancer treatment, facilitating precise diagnoses and enhancing treatment efficacy. The development and evaluation of image processing algorithms for medical image analysis heavily rely on the availability of medical images. In this study, we utilize dermoscopic images sourced from the PH2 database for analysis. Our results demonstrate that the Skin Cancer Classification (SCC) system, employing Convolutional Neural Networks (CNN), outperforms traditional methods in terms of accuracy (98.5%), specificity (100%), and sensitivity (97.5%) across both stages of evaluation. From the analysis, it is evident that the superiority of CNN-based SCC systems in accurately diagnosing skin cancer. This paper underscores the significance of leveraging advanced image processing techniques for medical image analysis, paving the way for reliable skin cancer classification systems with potential clinical applicability.