Related Experiment Video
Updated: May 28, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Unsupervised Anomaly Detection in Medical Imaging: A Survey of Methods, Challenges, and Future Directions
Boyang Liu1, Guangli Li1, Yuxing Zou1
1School of Information and Software Engineering, East China Jiaotong University, Nanchang 330013, China.
Bioengineering (Basel, Switzerland)
|May 27, 2026
Summary
Unsupervised anomaly detection in medical imaging helps find unusual patterns without needing abnormal examples. This review categorizes methods, discusses challenges like semantic modeling, and suggests future research for better early disease screening.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Unsupervised anomaly detection (UAD) in medical imaging identifies deviations from normal patterns without abnormal sample annotations.
- It is crucial for early disease screening, discovering unknown anomalies, and handling label-scarce or open-set scenarios.
- Medical images present unique challenges: complex anatomy, high semantic complexity of abnormalities, inter-individual variability, and high annotation costs.
Purpose of the Study:
- To systematically review unsupervised anomaly detection methods for medical imaging.
- To analyze 149 studies and 16 datasets, integrating task definitions, technological evolution, and clinical needs.
- To categorize methods, discuss challenges, and propose future research directions.
Main Methods:
- Categorization of mainstream UAD methods into four groups: image reconstruction-based, feature embedding-based, self-supervised learning-based, and foundation model-based.
- Systematic discussion of technical characteristics, applicable scenarios, and limitations of each category.
- Analysis of challenges including abnormal semantic modeling, anomaly quantification reliability, cross-center generalization, and evaluation protocols.
Main Results:
- The review provides a comprehensive overview of the current landscape of unsupervised anomaly detection in medical imaging.
- Key challenges are identified, including modeling complex abnormal semantics and ensuring reliable quantification and generalization.
- Future research directions such as multi-task modeling, incorporating medical priors, and multimodal fusion are highlighted.
Conclusions:
- Unsupervised anomaly detection methods are vital for advancing medical imaging analysis, especially in data-limited situations.
- Addressing the identified challenges is crucial for developing more robust and clinically applicable UAD systems.
- Future research should focus on advanced techniques like multi-task learning and multimodal fusion to enhance detection capabilities.