ISIC 2024 Medical Imaging Challenge

The ISIC (International Skin Imaging Collaboration) 2024 challenge involved classifying skin lesions from dermoscopic images — a high-stakes medical imaging task where misclassification directly impacts patient care. We placed 178th globally out of 2,739 teams, earning a Kaggle Bronze Medal.

Our approach focused on fine-tuning convolutional neural networks pre-trained on ImageNet, careful data augmentation to account for the relatively small dataset of labeled lesions, and ensemble methods combining predictions across models. The project emphasized the practical constraints of medical imaging: working with imbalanced classes (certain lesion types are far rarer than others), understanding failure modes at deployment, and the difference between high accuracy in validation versus robustness in clinical use.