Atlas is Your Perfect Context: One-Shot Customization for Generalizable Foundational Medical Image Segmentation

2025-12-19

AI Summary

Existing interactive foundation models (e.g., nnInteractive) suffer from poor generalization in rare clinical scenarios and rely heavily on labor-intensive, expert-crafted prompts, failing to meet the high-precision demands of medical image segmentation. To address this, we propose a novel framework featuring an atlas-guided, context-aware prompt generation mechanism and a test-time dual-path prediction fusion adapter—enabling zero-shot, single-instance-label-driven model customization without fine-tuning. Our method constructs an anatomical atlas via image registration to generate robust, anatomy-informed prompts, and integrates a lightweight adapter with multimodal foundation models for collaborative inference. Evaluated on a multicenter, multimodal, multi-organ dataset, our approach significantly improves segmentation accuracy—especially for small structures—while maintaining low deployment overhead and compatibility with real-time clinical workflows. Key innovations include the first atlas-prompt joint modeling framework and a fine-tuning-free, test-time adaptive mechanism.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Prompt Engineering / Prompting

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Accurate medical image segmentation is essential for clinical diagnosis and treatment planning. While recent interactive foundation models (e.g., nnInteractive) enhance generalization through large-scale multimodal pretraining, they still depend on precise prompts and often perform below expectations in contexts that are underrepresented in their training data. We present AtlasSegFM, an atlas-guided framework that customizes available foundation models to clinical contexts with a single annotated example. The core innovations are: 1) a pipeline that provides context-aware prompts for foundation models via registration between a context atlas and query images, and 2) a test-time adapter to fuse predictions from both atlas registration and the foundation model. Extensive experiments across public and in-house datasets spanning multiple modalities and organs demonstrate that AtlasSegFM consistently improves segmentation, particularly for small, delicate structures. AtlasSegFM provides a lightweight, deployable solution one-shot customization of foundation models in real-world clinical workflows. The code will be made publicly available.

Problem

• Customizes foundation models for medical image segmentation with one example.
• Improves segmentation of small structures using atlas-guided context prompts.
• Enhances generalization in underrepresented clinical contexts via test-time adaptation.

Innovation

• Atlas-guided framework for one-shot customization
• Context-aware prompts via atlas-query image registration
• Test-time adapter fusing atlas and foundation model predictions