In-Context Reverse Classification Accuracy: Efficient Estimation of Segmentation Quality without Ground-Truth

2025-03-06

AI Summary

Medical image segmentation faces clinical challenges due to the scarcity of ground-truth annotations and the difficulty of automatic, unsupervised quality assessment. To address this, we propose In-Context RCA, an unsupervised evaluation framework that requires no gold-standard labels. Our method integrates in-context learning with a retrieval-augmented selection mechanism, enabling reliable segmentation quality estimation from only a small set of reference images. We introduce the novel “Reverse Classification Accuracy” (RCA) metric, which quantifies segmentation confidence as the ability to reconstruct—or inversely classify—the original input image from its segmentation output. To our knowledge, this is the first work to synergistically combine in-context learning and retrieval augmentation for unsupervised segmentation quality assessment. Extensive experiments across multi-modal medical imaging datasets demonstrate high computational efficiency, strong robustness, and real-time capability for clinical quality control. The source code is publicly available.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Assessing the quality of automatic image segmentation is crucial in clinical practice, but often very challenging due to the limited availability of ground truth annotations. In this paper, we introduce In-Context Reverse Classification Accuracy (In-Context RCA), a novel framework for automatically estimating segmentation quality in the absence of ground-truth annotations. By leveraging recent in-context learning segmentation models and incorporating retrieval-augmentation techniques to select the most relevant reference images, our approach enables efficient quality estimation with minimal reference data. Validated across diverse medical imaging modalities, our method demonstrates robust performance and computational efficiency, offering a promising solution for automated quality control in clinical workflows, where fast and reliable segmentation assessment is essential. The code is available at https://github.com/mcosarinsky/In-Context-RCA.

Problem

• Estimates segmentation quality without ground-truth annotations.
• Uses in-context learning and retrieval-augmentation techniques.
• Provides efficient quality control for clinical workflows.

Innovation

• In-Context RCA estimates segmentation quality without ground truth.
• Uses in-context learning and retrieval-augmentation for reference selection.
• Efficient, robust quality estimation across medical imaging modalities.