MSAL-MIR: Multi-Stage Adaptive Loss for Medical Image Retrieval

Efficient and accurate retrieval of medical images underpins timely diagnosis and informed clinical decisions. This work introduces a novel multi-stage training paradigm designed for medical image retrieval. In the first stage, a ConvNeXt model pretrained on ImageNet is fine-tuned using Focal Loss to address class imbalance. Building on this foundation, the feature space is refined with Triplet Margin Loss, where chosen sample triplets are used to enhance discriminative learning. Our approach further streamlines the retrieval process by applying Global Max Pooling, L2 normalization, and Principal Component Analysis (PCA) for dimensionality reduction, followed by integration with Facebook AI Similarity Search (FAISS) for efficient similarity search. Experiments on the ISIC 2017 and COVID-19 chest X-ray datasets demonstrate that the proposed method achieves significant improvements in evaluation metrics, including mean Average Precision at 5(mAP@5), Precision at 1 (P@1), and Precision at 5 (P@5)