AI-driven melanoma detection in New Zealand: A ResNet50-based approach

Melanoma is a significant public health challenge in New Zealand due to high levels of ultraviolet (UV) radiation. While early detection is critical for improving patient outcomes, achieving consistently high diagnostic accuracy remains challenging. In this study, we used a fine-tuned ResNet50 convolutional neural network (CNN) for automated melanoma detection using dermoscopic images. The model was trained on large, publicly available datasets (HAM10000 and ISIC) using state-of-the-art techniques, including hyperparameter optimization, data augmentation, batch normalization, and adaptive learning rates, demonstrating its potential effectiveness. This approach achieved an accuracy of 95.1% on benchmark datasets. When tested on a dataset comprising New Zealand patients, the model achieved an accuracy of 78.4%. This underscores the need for further fine-tuning to optimize performance for local populations. The proposed model demonstrates promising performance within the New Zealand healthcare context, but also emphasizes the need for validation across diverse geographic populations to ensure broader clinical applicability. • A fine-tuned ResNet50 CNN model is developed for accurate melanoma classification from dermoscopic images. • The model demonstrates strong generalization across both international and region-specific datasets. • A prototype web-based diagnostic system is implemented to showcase real-time clinical applicability.