<![CDATA[Melanoma is a life-threatening skin cancer that poses challenges in regions with limited access to specialized medical personnel, such as Papua, Indonesia. Early diagnosis is essential, but accurate detection is hindered by the scarcity of dermatologists. This study develops a melanoma detection system using computer vision, utilizing the VGG16 architecture enhanced with the Convolutional Block Attention Module (CBAM) and fine-tuning via transfer learning. The model was trained on a dataset comprising melanoma and non-melanoma images, with data augmentation to address class imbalance. The model achieved an accuracy of 91.25%, precision of 92.31%, recall of 90%, and an F1-score of 91.13%, demonstrating reliable performance in melanoma classification. High specificity (92.5%) indicates a low false positive rate, while sensitivity (90%) shows effective melanoma detection, though the 10% false negative rate requires improvement. Future enhancements include increasing sensitivity through weighted loss functions, optimizing classification thresholds, and performing external validation. Additionally, Grad-CAM is used for interpretability, and a web-based application is proposed to support healthcare practitioners, offering an accessible diagnostic tool for melanoma screening in resource-limited settings.]]>
Copyrights © 2025
