How Dataset Imbalances Shape Medical Image Retrieval Accuracy

This article explores challenges and innovations in medical image retrieval, focusing on dataset imbalance, organ size and shape biases, and recall accuracy interpretation. It highlights a novel application of ColBERT-inspired re-ranking, demonstrating its feasibility in refining CBIR results by incorporating context such as user behavior and medical relevance. While no strong link was found between anatomical region size and retrieval recall, the study opens new pathways for improving image retrieval systems, balancing computational costs, and enhancing real-world usability.