Abstracts

Rationale: The complicated internal architecture and surface morphology of the hippocampus make it extremely challenging to image clearly, particularly in the head and tail, requiring voxel dimensions much less than 1 mm in all three dimensions. Common approaches to hippocampal imaging include 2D acquisitions with a resolution of less than 0.5 mm in the coronal plane but relatively thick slices (2-3 mm), or 3D acquisitions with isotropic resolution at or near 1 mm (0.8-1.0), but to simply apply these approaches with higher resolution results in individual scan times that are in excess of 10 min, poor SNR, and susceptibility to movement artifacts. We previously developed a method we refer to as HR-MICRA that overcomes these challenges by repeatedly acquiring an ultra high resolution image (0.5x0.5x0.7 mm) in a short scan time (6.5 min) and co-registering and averaging the images to correct for movement while improving SNR. This was originally developed on a first generation clinical 3T scanner, and here we present the details of optimization of this sequence for a state-of-the-art Siemens Prisma scanner. Methods: Initial sequence parameters were taken from Human Connectome Lifespan Project T2-weighted SPACE sequence, which was then modified from a resolution of 0.8mm isotropic to 0.5 mm in the coronal plane and 0.7 in the AP direction. The following parameters were varied with the goal of obtaining the highest gray-white contrast to noise ratio (CNR) in the shortest time: TR, TE, turbo factor (TF), phase partial Fourier (PPF), slice partial Fourier (SPF), PAT factor, 3D PAT factor, and elliptical scanning. Performance variables assess were time of acquisition (TA), gray matter signal intensity (SGM), white matter signal intensity (SWM), and combined gray matter and white matter noise, from which CNR and CNR efficiency (CNR/sqrt(TA)) can be derived. Because the HR-MICRA approach involves acquiring multiple sequential scans, a scan with a lower CNR may be preferable to one with a higher CNR that takes significantly longer to acquire due to the fact that the longer TA reduces the number of scans to be acquired in a fixed protocol duration. All tests were conducted on a single healthy volunteer using a 64-channel coil. Estimates of signal and noise were derived by acquiring two identical 3D image volumes for each sequence. Signal estimates were calculated as the average of all voxels in the designated region across both scans, and noise estimates were calculated from a difference image volume generated by subtracting the second image volume of the pair from the first. Results: The table shows the sequence parameters tested and results. The sequences in Group A represent sampling schemes with variable degrees of partial Fourier sampling and parallel reconstruction acceleration (PAT). The sequences in Group B are based on the best sequence from Group A and explore the effect of varying TR. The sequences in Group C explore the effects of varying TE. Conclusions: In Group A, as expected Sequence 1 and 2 have the highest CNR due to the more complete k-space sampling, but Sequence 4 has only slightly less CNR but a shorter scan time of 6:26 min, resulting in the optimum CNR efficiency. In Group B, shorter TRs resulted in significantly lower signal and consequently CNR and CNR efficiency. Increasing the TR to 3600 (Sequence 12) improved CNR and CNR efficiency but pushed scan time to over 7 min, which is considered undesirable in regard to minimizing intra-scan movement. In Group C, TE was reduced, revealing that a ~10% decrease in TE increased CNR, but further reductions in TE resulted in worsening CNR. Therefore Sequence 13 was considered the optimal sequence. Figure 1 shows an example of a single scan and Figure 2 shows an average of 6 coregistered scans. Funding: NIH/NINDS R01NS094743