Title:  P3382 Standard for Performance Metrics for Magnetic Resonance Image Reconstruction

Scope:  This standard establishes metrics to assess the quality of images reconstructed from undersampled frequency-wavenumber (k-space) measurements of static objects using
magnetic resonance imaging (MRI) techniques.  Since MRI data from patients is private information, this standard uses physical artifacts known as phantoms. The standard defines:
• quantitative metrics, such as feature contrast, aliasing artifacts, noise level, blurring, or peak signal-to-noise ratio, that can be used to determine the quality of MRI images;
• computational metrics for reconstruction algorithms, such as latency and throughput;
• an analysis method of imaging performance with respect to the defined metrics for available MRI reconstruction algorithms such as iterative compressed sensing techniques or deep learning with and without unrolling;
• a method to assess the training data required for deep learning approaches;
• a method to assess the impact of the sampling geometry (e.g. Cartesian) on MRI reconstruction performance.