Digital reference objects for evaluating algorithm performance in MR image formation.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Objective evaluation of MRI image quality (IQ) is crucial for algorithm development.
• Standardized metrics are needed for reproducible IQ assessment.
• Digital Reference Objects (DROs) offer a potential solution for objective IQ evaluation.

Purpose of the Study:

• To introduce and demonstrate the utility of Digital Reference Objects (DROs) for evaluating MRI image quality.
• To establish a basis for automated and reproducible IQ metrics in MR image formation.
• To facilitate the comparison of different MR image reconstruction algorithms.

Main Methods:

• Digital Reference Objects (DROs) were sampled directly in k-space using analytical Fourier transform formulas.
• A Convolutional Neural Network (CNN)-based denoising algorithm was applied to noisy ACR phantom images.
• Images were reconstructed from both measured and simulated k-space data for comparison.

Main Results:

• The CNN-based denoising algorithm produced virtually identical results on both measured and simulated ACR phantom data.
• Visual and quantitative comparisons confirmed the consistency of the denoising performance.
• This demonstrates the fidelity of DROs in representing real-world phantom data.

Conclusions:

• Digital Reference Objects (DROs) can guide technology selection in developing new MR image formation algorithms, including deep learning approaches.
• The use of DROs represents a significant step towards achieving reproducible MR image formation.
• This methodology supports the standardization of MRI IQ assessment.