Correcting synthetic MRI contrast-weighted images using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Quantitative MRI

Background:

• Synthetic MRI enables rapid generation of diverse image contrasts from quantitative parameter maps.
• Current synthetic MRI methods often exhibit contrast deviations from experimental scans due to unmodeled physical effects.
• These unmodeled effects, including diffusion and susceptibility, limit the accuracy of synthetic images.

Purpose of the Study:

• To develop and validate a novel deep learning approach for correcting synthetic MRI contrasts.
• To improve the fidelity of synthetic MRI by accounting for unmodeled physical effects.
• To enhance the clinical utility of synthetic MRI by reducing contrast mismatch with experimental scans.

Main Methods:

• A physics-informed deep learning model was proposed to generate a multiplicative correction term.
• This correction term addresses unmodeled physical effects impacting synthetic image contrast.
• The method was validated on synthesizing inversion recovery fast spin-echo sequences using a 2D multi-contrast MRI acquisition.

Main Results:

• The proposed deep learning correction visually and numerically reduced contrast mismatch compared to conventional synthetic MRI.
• The method demonstrated improved accuracy in synthesizing arbitrary inversion recovery fast spin-echo contrasts.
• A preliminary reader study indicated statistically significant improvements in contrast and signal-to-noise ratio (SNR) over standard synthetic MRI.

Conclusions:

• The novel deep learning approach effectively corrects unmodeled physical effects in synthetic MRI.
• This method enhances the accuracy and diagnostic quality of retrospectively synthesized MRI contrasts.
• The findings suggest a significant advancement in the practical application of synthetic MRI for clinical imaging.