SynPo: Boosting Training-Free Few-Shot Medical Segmentation via High-Quality Negative Prompts

2025-06-18

AI Summary

Existing zero-shot large vision models (LVMs) for few-shot segmentation of low-contrast medical images suffer from neglecting negative prompts and poor generalization. To address this, we propose SynPo—a training-free segmentation framework. Its core contributions are: (1) a novel Confidence Map Coordination Module that fuses multi-scale features from DINOv2 and SAM; and (2) a joint positive–negative point prompt optimization strategy, leveraging Gaussian sampling and independent K-means clustering to enhance the quality and discriminability of negative prompts. SynPo significantly improves SAM’s boundary discrimination and sensitivity to small structures in zero-shot settings. Evaluated on multi-organ medical datasets, SynPo achieves performance on par with state-of-the-art supervised few-shot methods—yielding substantial gains in boundary accuracy (+3.7%) and small-object recall (+5.2%). This work establishes a new paradigm for training-free medical image segmentation.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

The advent of Large Vision Models (LVMs) offers new opportunities for few-shot medical image segmentation. However, existing training-free methods based on LVMs fail to effectively utilize negative prompts, leading to poor performance on low-contrast medical images. To address this issue, we propose SynPo, a training-free few-shot method based on LVMs (e.g., SAM), with the core insight: improving the quality of negative prompts. To select point prompts in a more reliable confidence map, we design a novel Confidence Map Synergy Module by combining the strengths of DINOv2 and SAM. Based on the confidence map, we select the top-k pixels as the positive points set and choose the negative points set using a Gaussian distribution, followed by independent K-means clustering for both sets. Then, these selected points are leveraged as high-quality prompts for SAM to get the segmentation results. Extensive experiments demonstrate that SynPo achieves performance comparable to state-of-the-art training-based few-shot methods.

Problem

• Improving negative prompts for medical image segmentation
• Enhancing low-contrast medical image segmentation accuracy
• Training-free few-shot method using LVMs like SAM

Innovation

• Uses Confidence Map Synergy Module
• Selects high-quality prompts via K-means
• Leverages DINOv2 and SAM strengths