Evaluating Machine Learning-Based MRI Reconstruction Using Digital Image Quality Phantoms.

Area of Science:

• Medical Imaging
• Machine Learning
• Image Reconstruction

Background:

• Quantitative evaluation is crucial for machine learning (ML)-based magnetic resonance imaging (MRI) reconstruction.
• Existing metrics like MSE, SSIM, and PSNR inadequately assess clinical image quality.

Purpose of the Study:

• To develop and validate a novel pipeline for evaluating ML-based MRI reconstruction using digital image quality phantoms.
• To establish automated methods for assessing key clinical image quality parameters.

Main Methods:

• Created digital k-space phantoms simulating the ACR large physical phantom with variable parameters (size, SNR, resolution, contrast).
• Developed an evaluation pipeline incorporating metrics for geometric accuracy, uniformity, ghosting, sharpness, SNR, resolution, and low-contrast detectability.
• Assessed an ML reconstruction model using the proposed pipeline across different training scenarios.

Main Results:

• The proposed digital phantom and automated evaluation pipeline effectively assess ML-based MRI reconstruction.
• Training data with lower undersampling factors and larger coil coverage resulted in improved model performance.
• Identified key factors influencing ML reconstruction quality.

Conclusions:

• The developed comprehensive and standardized pipeline enhances understanding of ML reconstruction performance.
• This methodology can guide future development and advancement of ML algorithms for MRI.
• Digital phantoms offer a flexible and reproducible alternative for MRI quality assessment.