An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and...

Area of Science:

• Medical Imaging Analysis
• Machine Learning in Healthcare
• Neuroimaging Quality Control

Background:

• Medical imaging datasets are growing, but ensuring sample quality and identifying artefacts remains challenging.
• Existing automatic artefact detection methods often require extensive training data, which is scarce for rare artefacts.
• This limitation hinders the development and clinical application of machine learning for medical image analysis.

Purpose of the Study:

• To develop a novel framework for robust artefact detection in brain MRI scans.
• To address the challenge of limited labelled data for training artefact detection models.
• To improve the quality control of medical imaging datasets for clinical research and applications.

Main Methods:

• Utilized physics-based artefact generators for data augmentation, creating synthetic brain MRI scans with controlled artefacts.
• Developed a comprehensive set of abstract and engineered image features for compact image representation.
• Implemented an artefact-specific feature selection process to optimize classification performance.
• Employed Support Vector Machine (SVM) classifiers for artefact identification.

Main Results:

• The proposed framework significantly outperforms traditional methods in artefact detection accuracy and efficiency.
• Data augmentation improved performance metrics (accuracy, F1, F2, precision, recall) by up to 12.5 percentage points.
• The pipeline processes single scans in under a second, enabling potential real-time deployment.
• Validated on datasets with both synthetic and real artefacts from a multiple sclerosis clinical trial.

Conclusions:

• The novel framework effectively addresses the scarcity of labelled artefact data through physics-based augmentation.
• The system provides a low-computation cost, high-performance solution for brain MRI quality control.
• This approach facilitates the development of automated quality control systems for high-throughput clinical applications.