Decoupled Single-Mask Annotation Noise Detection via Cross-Sectional Patch Self-Consistency

2026-07-07

AI Summary

This study addresses the challenge of auditing annotation noise in single-annotator vascular CT segmentation, where the absence of multi-annotator validation and the coupling of existing noise-handling methods with model training hinder reliable error detection. To overcome this, the authors propose a decoupled noise detection framework that leverages the repetitive appearance patterns of vessel cross-sectional image patches across anatomical locations and subjects. By employing vector retrieval to identify intensity-similar neighboring patches, the method quantifies mask inconsistency among these neighbors to produce interpretable, region-level evidence of annotation noise—without requiring integration into model training. The approach reveals that annotation error rates are strongly associated with vessel orientation, cross-sectional area, and intensity, with transverse and oblique vessels exhibiting 5.1 times higher error rates than axial structures. Experiments on coronary CT angiography data demonstrate the framework’s effectiveness and its ability to enhance downstream model robustness.

Abstract

Vascular computed tomography datasets are commonly annotated only once per scan, yielding the pervasive yet under addressed problem of single mask annotation noise. Existing solutions either require costly multirater fusion or are coupled with network training, preventing explicit auditing of where and why labels fail. We introduce a decoupled framework for single-mask annotation noise detection that leverages cross-sectional patch self-consistency to produce interpretable and auditable noise evidence. Tubular anatomy exhibits strong cross-sectional recurrence: patches extracted orthogonally along vessel centrelines recur in appearance across locations and subjects. Thus, anatomically similar patches should have consistent masks, and disagreement signals unreliable annotation. Our method samples cross-sectional patches, retrieves intensity-equivalent neighbours via scalable vector search, and computes a patch-level noise score from statistical mask disagreement, yielding explicit image-mask evidence for every flagged region. Aggregating scores produces scan-level quality maps for dataset quality assessment or quality-weighted training. Experiments on the coronary CT dataset validate the detected noise for improving training robustness and reveal systematic annotation biases. Specifically, transverse and oblique vessels exhibit 5.1 times higher error rates than axis-aligned structures, with additional correlations to cross-sectional area and intensity. Code is available here.

Problem

• annotation noise
• single-mask annotation
• vascular CT
• label quality
• medical image annotation

Innovation

• annotation noise detection
• cross-sectional patch self-consistency
• decoupled framework
• vascular CT
• interpretable auditing