Unsupervised Anomaly Detection for Image Dataset Quality Assurance in Multi-Center Breast MRI

2026-08-17

AI Summary

This study addresses quality control challenges in multi-center breast MRI by proposing an unsupervised anomaly detection framework. Through a controlled benchmark encompassing 17 anomaly types, we identify near out-of-distribution (OOD) detection as a critical bottleneck. To overcome this, we introduce a novel approach integrating projection methods, 3D reconstruction, and hybrid OOD detection, enhanced with domain-specific features and positional encoding for precise identification. Experiments demonstrate that the projection method achieves an AUROC of 0.954, while 3D reconstruction exhibits superior generalization, underscoring the necessity of domain adaptation. This research establishes a scalable, automated quality control paradigm, providing essential benchmarks and methodological support for data safety in medical AI, although detecting implants remains challenging.

Abstract

Corrupted, inconsistent, or anomalous data silently threatens the safety and reliability of medical AI. Despite growing regulatory recognition of dataset quality assurance (QA) for high-risk medical AI, scalable automated detection remains underdeveloped. We employ unsupervised anomaly detection (AD) and out-of-distribution (OOD) detection as an automated dataset QA mechanism for multi-center dynamic contrast-enhanced breast MRI. We build a controlled AD benchmark of 17 realistic QA-relevant anomaly types from six public datasets (protocol violations, processing errors, incorrect anatomical regions) and propose a taxonomy of radiological image anomalies based on human visual perception, enabling fine-grained analysis of AD failure modes. The benchmark includes near-, medium-far-, far-OOD samples, as well as in-distribution and external normal data. Four methods are evaluated: a projection-based method extended with a domain-specific feature extractor and a novel positional encoding, a reconstruction-based approach extended to full 3D volumes with an augmented training objective, and two unmodified hybrid OOD detection methods. Medium-far- and far-OOD samples are detected reliably, whereas near-OOD samples and external normal data from unseen institutions expose method-specific differences. The 3D reconstruction-based approach best balances detection performance (AUROC: 0.936) and generalization to unseen institutions. The projection-based method with positional encoding achieves the highest overall detection performance (AUROC: 0.954). Both hybrid methods exhibit critical failure modes, confirming that methods validated for one modality or anatomy may not generalize without domain-specific adaptation. Implants and mastectomies remain an open challenge for all methods. Our results establish a foundation and practical guidance on scalable unsupervised QA in medical AI pipelines.

Problem

• Unsupervised Anomaly Detection
• Dataset Quality Assurance
• Breast MRI
• Out-of-Distribution Detection
• Medical AI

Innovation

• Unsupervised Anomaly Detection
• Dataset Quality Assurance
• Positional Encoding
• 3D Reconstruction
• OOD Detection Benchmark