Rapid High Resolution T1-Weighted Hippocampus Imaging with Yarn-Ball Acquisition
Rob Stobbe1, Peter Seres1, Don Gross1, and Christian Beaulieu1

1University of Alberta, Edmonton, AB, Canada


For spoiled steady-state T1-weighted imaging, readout duration (TRO) and repetition time (TR) increase result in greater contrast-to-noise ratio (CNR) efficiency. Novel 3D-twisting Yarnball acquisition realizes this advantage without scan-time penalty (more of k-space sampled with increased TRO), but increased TRO results in greater point-spread-function smearing. Following TRO optimization, Yarnball is used to produce whole-brain 0.36x0.36x1.08 mm3 coronal (defined by 1/(2kmax)) images in 10 minutes (with 2 averages). Compared to 3D-MP-RAGE (same scan time and voxel volume) Yarnball images have greater resolution and grey-white CNR, facilitating sharper depiction of internal hippocampus architecture.


High resolution imaging is required to distinguish hippocampal architecture, but this can be time consuming. Here, novel 3D-twisting Yarnball1 is considered as a T1-weighted k-space acquisition alternative to previously considered 3D-MP-RAGE with voxels of 0.38x0.38x0.8 mm3 and 0.6x0.6x0.6 mm3 and long scan times of 37 minutes2 and 33 minutes3 respectively at 3T. Following readout duration (TRO) optimization for contrast-to-noise ratio (CNR) and resolution, 0.36x0.36x1.08 mm3 (defined by 1/(2kmax)) Yarnball was compared to 0.45x0.45x1.35 mm3 MP-RAGE (1/kvolume=0.27 mm3 for both) for patient friendly 10 minute scanning of the whole brain at 3T.


Simple T1-based spoiled steady-state simulation suggests grey-white CNR efficiency increase with longer TRO and TR (Figure 1), but while longer TRO and TR are constrained by scan duration for GRE, that is not the case for Yarnball1 (see Figure 2 for an example gradient waveform and k-space trajectory) which exhibits even greater sampling efficiency with longer readouts (more of k-space sampled per acquisition). For Yarnball, TRO and TR can be arbitrarily increased to maximize CNR efficiency without scan time penalty. However, longer readouts result in greater point-spread-function (PSF) smearing, a consequence of both T2* decay and off-resonance.

To aid TRO selection for high-resolution hippocampus imaging, TRO = 6, 10, 14, and 20 ms Yarnball waveforms fully sampling the human head were tested with TR = 11, 16, 20, and 25 ms, prescribed flip-angle = 12o, 14o, 16o, and 18o, and number of trajectories = 9537, 6694, 5211, and 4260. In each case voxels were 0.63x0.63x0.95 mm3 coronal (defined by 1/(2kmax)) and scan time was 1:45 minutes. Images were acquired on a Siemens 3T Prisma (64-channel coil) from three healthy volunteers (41, 24, 23 years). All images included 1-1 water excitation and were reconstructed with a β = 2 Kaiser filter. Relative (between scan) CNR was assessed between the inferior longitudinal fasciculus and the collateral sulcus. Images were also acquired from a resolution phantom on-resonance and 20 Hz off-resonance (selected from Bo mapping observation in the hippocampus). Further analysis included calculation of the PSF Full-Width and Full-Volume Half-Max (FWHM/FVHM) with T2*=45 ms both on-resonance and 20 Hz off-resonance.

From the experiment above (and explained in Results), TRO=15 ms was selected for testing high-resolution 3D-T1-weighted Yarnball imaging in vivo (Figure 2 pulse sequence). With TR=20 ms, a total of 15801 trajectories provided full head sampling for 0.36x0.36x1.08 mm3 (coronal) voxels in only 5:19 min. Two sequential images were acquired and realigned before averaging to reduce motion-related smearing (total 10:38 min). Note that 1/kvolume=0.27 mm3. Voxel anisotropy was chosen to aid in-plane hippocampus architecture detection. For comparison, 0.45x0.45x1.35=0.27 mm3 3D-MP-RAGE was acquired coronally in 10:51 minutes: FoV=216x183x238 mm3, matrix=406x480x176, BW=200 Hz/px, flip-angle=11o, TI=810 ms, TR=1600 ms, no under-sampling. Sequence parameters were selected from T1-based simulation, and images with and without Siemens (medium, edge-enhancement 3, smoothing 2) filtering compared. High resolution images were acquired from a male volunteer (25 years).


Representative human brain (zoomed-in on hippocampus body) and resolution phantom images are shown in Figure 3 for the four different readout durations. Grey-white CNR increases 80% from TRO=6 ms to 20 ms, similar to simulation (Figure 1). For on-resonance, 0.75 mm resolution element distinction remains similar in the phantom from TRO=6 ms to 20 ms. This is predicted from calculated FWHM and FVHM measures that increase by only 2% and 7%, respectively, over this TRO range. However, resolution element distinction is worsened to 1.0 mm for TRO=20 ms when 20 Hz off-resonance; calculated FWHM and FVHM measures increase by 6% and 22% respectively over TRO=6 ms. Given the detrimental effect of off-resonance, TRO was limited to 15 ms for high-resolution imaging. Yarnball offers both higher CNR and superior resolution element distinction when compared with MP-RAGE (Figure 4). Zoomed-in slices of the right hippocampus more clearly depict the stratum lacunosum-moleculare (SLM) on Yarnball than either unfiltered (very noisy) or Siemens-filtered MP-RAGE images (Figure 5). Note that the SLM is often used to help define hippocampal subfields4.

Discussion and Conclusion

Efficient Yarnball acquisition facilitates the use of longer TRO and TR for greater steady-state T1-weighted CNR efficiency without increasing scan duration. However, practical TRO is limited by off-resonance related PSF smearing. Robust shimming over the hippocampus may improve Yarnball utility in this regard. Comparing patient friendly 10 minute scans, Yarnball provides superior images of the hippocampus in terms of both CNR and resolution and enables greater internal hippocampus architecture distinction than MP-RAGE.


Canadian Institutes of Health Research


1. Stobbe RW, Beaulieu C. Rapid 3D spoiled steady-state imaging with yarn-ball acquisition. ISMRM, Abstract 2442. 2015 (Toronto).

2. Van Leemput K, Bakkour A, et al. Model-based segmentation of hippocampal subfields in ultra-high resolution in vivo MRI. Med Image Comput Comput Assist Interv. 2008;11(Pt1):235-243.

3. Kaluga-Yoskovitz J, Bernhardt BC. Multi-contrast submillimetric 3 Tesla hippocampal subfield segmentation protocol and dataset. Sci Data. 2015;2:150059.

4. Steve TA, Yasuda CL et al. Development of a histologically validated segmentation protocol for the hippocampal body. Neuroimage. 2017;157:219-232.

5. Stobbe RW, Beaulieu C. Assessment of averaging spatially correlated noise for 3D radial imaging. IEEE Trans Med Imag. 2011;30(7):1381-1390.


Figure 1: T1 based simulation shows that the relative grey-white matter CNR-efficiency of spoiled steady-state imaging increases with readout duration (TRO) and repetition time (TR). Here a non-acquisition time (including excitation and spoiling) of 5 ms results in TR ranging from 10 to 25 ms. CNR-efficiency increase is the result of both increased sampling duty-cycle and the facilitation of larger flip-angles within the steady-state regime. T1 values of 830 ms (grey) and 1330 ms (white) were used for this simulation. Relative CNR efficiency of 1.0 corresponds to the shortest TRO = 6 ms used in this study.

Figure 2: The high-resolution 3D T1-weighted Yarnball sequence where the gradient waveform shown prescribes the corresponding trajectory through k-space. The readout duration of 15 ms for high-resolution imaging was chosen from the optimization performed and shown in Figure 3.

Figure 3: Readout duration increase from 6 ms to 20 ms yields 80% greater grey-white matter CNR for the same 1:45 minute scan time and 0.63x0.63x0.95 mm3 voxels (defined by 1/(2kmax)), as determined from three volunteers (mean +/- standard deviation). When on-resonance, resolution element distinction and both FWHM and FVHM remain similar for readouts from 6 ms to 20 ms. However, when 20 Hz off-resonance, as may be the case in the hippocampus (in-house observation at 3T), the PSF and resolution element signal is particularly smeared for the 20 ms readout case.

Figure 4: Coronal slices from (a) Yarnball (0.36x0.36x1.08 mm3, 1/(kvolume=0.27 mm3) and (b) Siemens-filtered MP-RAGE (0.45x0.45x1.35=0.27 mm3) images acquired in 10:38 and 10:51 minutes, respectively. B1-inhomogeneity correction (N4 software) was applied equally to both Yarnball and MP-RAGE. Grey-white matter CNR in the vicinity of the hippocampus is ~7 for Yarnball and ~6 for MP-RAGE. Resolution phantom (top right) element distinction down to 0.5 mm is superior for the Yarnball images.

Figure 5: Zoomed-in coronal slices of the right hippocampus for (a) Yarnball, (b) MP-RAGE (no filtering) and (c) MP-RAGE (Siemens filtering). In general, boundaries are more sharply defined with Yarnball, a function of both spherical sampling to greater k-space values than MP-RAGE, and noise power spectral density in which low frequency noise is greatly attenuated (i.e. smaller noise ‘speckles’). The latter is a product of oversampling around the centre of k-space5. As a result, the stratum-lacunosum-moleculare (SLM) is better defined on the Yarnball images and is seen as the white curved line.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)