Intelligent Melanoma Detection in Dermoscopic Images Using Deep Learning and Class Activation Maps (Grad-CAM) to Enhance Clinician Trust

Melanoma is one of the most dangerous forms of skin cancer, and its early detection plays a vital role in patient survival. However, accurately diagnosing skin lesions remains challenging even for dermatologists due to the high visual similarity between benign and malignant moles.In this research, we utilized a Convolutional Neural Network (CNN) with an advanced EfficientNet-B۰ architecture, trained using Transfer Learning on the standard ISIC datasets. To address the “black box” challenge inherent in deep learning models, we employed the Grad-CAM technique to generate heatmaps, visualizing the regions within the images that most significantly influenced the model’s decision-making process.Our proposed model achieved an accuracy of ۹۲% and a sensitivity of ۸۹% in lesion classification. The Grad-CAM results demonstrated that the model effectively focuses on clinical features, such as asymmetry and irregular lesion borders, which aligns with the diagnostic criteria used by dermatologists.The findings of this study suggest that combining deep neural networks with interpretability tools can serve as a robust Clinical Decision Support System (CDSS). By reducing diagnostic errors, this approach can significantly contribute to faster and more accurate melanoma screening.

کلیدواژه ها:

Deep Learning ، Melanoma ، Medical Diagnosis ، Explainable AI (XAI) ، Dermoscopic Imaging