Multi-AD: cross-domain unsupervised anomaly detection for medical and industrial applications

2025-09-01

Pattern Recognition

AI Summary

This work addresses the challenge of unsupervised cross-domain anomaly detection in medical and industrial imaging, where the absence of annotated anomalies hinders model performance. To this end, the authors propose Multi-AD, a novel framework built upon a teacher–student architecture that integrates multi-scale features and incorporates a channel attention mechanism. By synergistically combining knowledge distillation with a discriminator network, Multi-AD significantly enhances the detection of subtle and multi-scale anomalies while improving cross-domain generalization. Extensive experiments demonstrate state-of-the-art performance across multiple medical and industrial datasets, achieving image-level AUROC scores of 81.4% (medical) and 99.6% (industrial), and pixel-level AUROC scores of 97.0% and 98.4%, respectively—substantially outperforming existing methods.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Problem

• anomaly detection
• cross-domain
• unsupervised learning
• medical imaging
• industrial inspection

Innovation

• unsupervised anomaly detection
• cross-domain
• knowledge distillation
• squeeze-and-excitation
• multi-scale feature fusion