Diffusion tensor imaging using partial Fourier STEAM MRI with projection onto convex subsets reconstruction.

Diffusion-weighted single-shot STEAM MRI allows for diffusion mapping of the human brain without sensitivity to resonance offset effects. In order to compensate for its inherently lower SNR and speed than echo-planar imaging, this work describes the use of partial Fourier encoding in combination with image reconstruction by the projection onto convex subsets algorithm. The method overcomes phase distortions in diffusion-weighted partial Fourier acquisitions that disturb the conjugate complex symmetry of k-space and preclude the use of respective reconstruction techniques. In comparison with full Fourier encoding and a static flip angle for the STEAM readout pulses, experimental results at 2.9 T demonstrate a gain in relative SNR per unit time by 20% for 5/8 phase encoding with optimized variable flip angles. Simultaneously, the imaging time is reduced from about 670 ms (80 echoes) to 440 ms (50 echoes). Current implementations at 2 x 2 mm2 in-plane resolution comprise a protocol for clinical anisotropy studies (12 diffusion-encoding gradient directions at 1000 s mm(-2)) covering 18 sections of 4-mm thickness within a measurement time of 8.5 min (5 averages) and a version optimized for fiber tracking using 24 gradient directions and 38 sections of 2-mm thickness yielding a measurement time of 29.5 min (4 averages).