Two-Stage Fine-Tuning of ResNet50 for High-Sensitivity Melanoma Detection on Dermoscopic Images

Melanoma is the most dangerous form of skin cancer, and catching it early significantly improves patient outcomes. Five-year survival rates exceed 99% when detected at an early stage but fall sharply once the cancer spreads. This paper proposes and evaluates a two-stage fine-tuning approach for ResNet50 applied to binary melanoma classification on dermoscopic images. The core challenges addressed are class imbalance and suboptimal transfer learning. Random oversampling was used to achieve a 1:1 class balance on the HAM10000 dataset. Stage 1 trained only the classification head with the ResNet50 base frozen, while Stage 2 fine-tuned all layers jointly at a low learning rate of 1e-5 to prevent catastrophic forgetting. On an independent test set of 3,826 images, the model achieved an AUC-ROC of 0.9559, accuracy of 88.34%, sensitivity of 87.56%, and specificity of 89.13%. Code and a working Streamlit detection application are available at https://github.com/Aryanbhagat23/melanoma-detection