A review of deep learning-based Unsupervised Anomaly Detection in brain MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Manual assessment of brain MRI scans is time-consuming.
• Deep Learning requires large annotated datasets, which are often unavailable.
• Unsupervised Anomaly Detection (UAD) can identify abnormalities without pixel-level annotations by learning normal patterns.

Purpose of the Study:

• To review and systematically compare Unsupervised Anomaly Detection (UAD) approaches for brain MRI.
• To address the challenge of inconsistent evaluation contexts across different UAD studies.
• To provide a comprehensive resource for UAD in brain MRI analysis.

Main Methods:

• Literature review of Unsupervised Anomaly Detection (UAD) methods for brain MRI.
• Systematic collection and comparison of existing UAD approaches.
• Analysis of evaluation methodologies and contexts across studies.

Main Results:

• UAD has advanced significantly in brain MRI analysis.
• Inconsistent evaluation contexts (acquisition parameters, preprocessing, scoring) hinder direct comparison of models.
• There is a clear need for standardized comparative studies in the context of MRI scans.

Conclusions:

• UAD is a promising technique for detecting unseen abnormalities in brain MRI.
• Standardized evaluation frameworks are crucial for assessing and advancing UAD methods in neuroimaging.
• This work provides a valuable collection of resources to facilitate future research.