Data-driven optimization of sampling patterns for MR brain T_1ρ mapping.

Area of Science:

• Medical Imaging
• Biophysics
• Data Science

Background:

• 3D T1ρ MRI is crucial for brain imaging but often requires long acquisition times.
• Compressed sensing (CS) reconstruction techniques can accelerate MRI by using undersampled data.
• Optimizing sampling patterns (SPs) is key to maximizing the effectiveness of CS reconstruction.

Purpose of the Study:

• To apply a data-driven optimization algorithm, bias-accelerated subset selection, to generate optimized sampling patterns (OSPs) for 3D T1ρ brain MRI.
• To evaluate the performance of OSPs with different compressed sensing reconstruction methods for accelerating 3D T1ρ mapping.

Main Methods:

• Five healthy volunteers underwent fully sampled 3D T1ρ MRI acquisition.
• Variable density (VD) and Poisson disc (PD) undersampling schemes were used as input for SP optimization.
• Three CS reconstruction methods (spatiotemporal finite differences, low-rank plus sparse, low-rank) were compared using normalized root mean square error (NRMSE).

Main Results:

• Optimized sampling patterns (VD-OSP and PD-OSP) significantly reduced NRMSE compared to non-optimized patterns (VD and PD) across various acceleration factors (AF).
• At AF=30, VD-OSP with spatiotemporal finite differences achieved NRMSE=0.078, outperforming non-optimized VD (NRMSE=0.087).
• At AF=4, VD-OSP (NRMSE=0.057) outperformed PD-OSP (NRMSE=0.060), demonstrating the superiority of optimized sampling.

Conclusions:

• Data-driven optimized sampling patterns, when combined with appropriate CS reconstruction, can effectively accelerate 3D T1ρ brain MRI.
• This approach holds promise for improving the efficiency of brain imaging applications using 3D T1ρ mapping.