Enhancing Few-Shot Medical Image Segmentation with Refined Prototypes

Abstract

Deep learning has achieved tremendous success in computer vision, while medical image segmentation (MIS) remains a challenge, due to the scarcity of data anno tations. Deep learning-based medical image segmentation methods are recognized as data-hungry techniques that require large amounts of data with manual annota tions. However, manual annotation is expensive in the field of medical image analysis, which requires domain-specific expertise. To address this challenge, few-shot learn ing has the potential to learn new classes from only a few labeled examples. In this work, we propose a framework for few-shot medical image segmentation based on cross-masked attention Transformer and Region-enhanced Prototypical Transformer. Our proposed network mines the correlations between the support image and query image, limiting them to focus only on useful foreground information and boosting the representation capacity of both the support prototype and query features. To mit igate the effects of large intra-class diversity, we further design a subdivision strategy to produce a collection of regional prototypes from the foreground of the support, and a self-selection mechanism is proposed to incorporate into the Bias-alleviated Trans former (BaT) block to suppress or remove interferences present in the query prototype and regional support prototypes. Thus, we have combined the feature enhancement and prototype generation in the Enhanced Prototype Generation EPG module which will iteratively update the generated query mask by taking the query mask gener ated in the previous iteration and finally produce more accurate global prototypes for Few-Shot Medical Image Segmentation. We conducted experiments on three publicly available medical image datasets, Abd-CT, Abd-MRI, and CMR to show the segmen tation results of the available state-of-the-art methods. Our experiments yielded final mean Dice scores of 50.82% on Abd-CT, 45.82% on Abd-MRI, and 75.08% on the CMR dataset, demonstrating competitive performance across these challenging datasets.