Two-Minute Deep Learning-Powered Brain Quantitative Mapping: Accelerating Clinical Imaging With Synthetic Magnetic...

Area of Science:

• Neuroimaging
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Quantitative magnetic resonance imaging (MRI) provides critical tissue property maps (T1, T2, PD) for disease diagnosis.
• Protracted scanning duration is a major limitation for clinical adoption of quantitative MRI.

Purpose of the Study:

• To evaluate deep learning (DL)-based superresolution reconstruction of ultrafast synthetic MRI.
• To determine if DL can generate accurate quantitative T1/T2/PD maps comparable to routine scans.
• To assess if scan time can be substantially reduced while preserving diagnostic image quality.

Main Methods:

• Prospective enrollment of 151 healthy adults and 7 patients.
• Dual-protocol MRI scans (routine and fast synthetic) on a 3.0T scanner.
• Application of a superresolution generative adversarial network (SRGAN) for image reconstruction.
• Validation using paired t-tests, Bland-Altman analysis, and coefficients of variation.

Main Results:

• Strong correlations between DL-reconstructed and reference T1/T2/PD values (R² > 0.97).
• Small average biases for T1 (0.93%), T2 (-0.85%), and PD (0.31%).
• SRGAN suppressed noise and artifacts, restoring image fidelity close to routine scan quality.

Conclusions:

• DL superresolution on ultrafast synthetic MRI yields whole-brain quantitative maps with strong correlation to routine scans.
• Acquisition time is halved with maintained diagnostic image quality.
• This approach shows promise for accelerating the clinical deployment of quantitative brain imaging.