INTRODUCTION

T_1ρ is used to study low frequency biological processes to which other contrast mechanisms are less sensitive and has therefore gained increasing interest in basic and clinical research of the brain. For example, it was shown to detect abnormalities in the white matter related to multiple sclerosis [1]. However, long acquisition times are required to obtain 3D T_1ρ maps with full-brain coverage and high isotropic resolution. Additionally, offline processing to produce relaxometry maps may limit its implementation in clinical workflows.
Here, we present a 3D T_1ρ mapping method with an accelerated compressed sensing acquisition, and inline map reconstruction. The first goal of this study was to investigate sequence and reconstruction parameters for a full-brain protocol with 1-mm isotropic resolution and acquisition times (TA) that are feasible for clinical applications. Second, we tested the optimized protocol for repeatability and reproducibility in healthy subjects.

METHODS

IRB-approved experiments were performed in nine healthy subjects (three females, 28.6±3.3yo) at 3 T (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) using a commercial 64-channel head coil. In each subject, whole-brain images were acquired with a segmented 3D FLASH [2] prototype sequence employing a Cartesian variable-density spiral-phyllotaxis sampling pattern [3] in the two phase-encoding directions. Readouts of the FLASH block followed a center-out radial-like trajectory (Figure 1A). A T_1ρ preparation module, with adiabatic 90° pulses [4] and a rectangular spin-lock pulse (SL) at 500 Hz, was played out before each readout train (Figure 1B). The SL pulse duration (SL-time) was changed to adjust the T_1ρ-weighting of the image. Four volumes were acquired consecutively in one scan with SL-times of 2, 30, 50 and 80 ms for T_1ρ mapping. Other protocol parameters included: TR=2000 ms, 160 readouts/TR, 8° FLASH RF-excitation angle, FLASH-TR/TE=5.8/1.9 ms, FOV=256x192x256 mm³, resolution=1x1x1 mm³, sagittal orientation.
Reconstructions of all scans were performed inline on the scanner with a prototype compressed sensing algorithm [5] using a Haar wavelet regularization. T_1ρ maps were automatically generated inline from the four SL-time volumes using a log-linear voxel-wise fit of $$$M_{t}=M_{0}\exp(-t/T_{1\rho})$$$, with t denoting the SL-time.
To optimize TA with respect to image quality, T_1ρ-prepared images with SL-time=80 ms were acquired in one subject with identical protocol parameters and acceleration factors (AF)=1/1.9/3.3/4/5.1/6.1/7/8. AF=1 was used as reference to compute NRMSE and SSIM [6] for the other AFs. T_1ρ maps (i.e. images with SL-time=2/30/50 ms) were acquired for AF=1.9-8 and evaluated visually by the authors.
After selection of an AF, the optimized T_1ρ mapping protocol was tested in eight other subjects for repeatability where a first Reference scan was compared to a second Repeat scan performed in the same subject after repositioning during the same session. For reproducibility a third scan was performed in the same subject one week later (One-Week scan).
Additionally, an MP-RAGE [7] sequence was acquired as anatomical reference for automatic segmentation of volumes-of-interest (VOI) before each T_1ρ-map scan.
Offline analysis: masks of nine bilateral VOIs (in white and gray matter, and deep brain structures) were obtained using the prototype segmentation software MorphoBox [8]. Subsequently, VOIs were rigidly registered [9] to the native space of the T_1ρ maps. Median T_1ρ values were extracted from each VOI, and their mean and standard deviation were computed across subjects for Reference, Repeat and One-Week scans. Linear regression and Bland-Altman analyses of the medians were performed for repeatability and reproducibility assessment [10].

RESULTS AND DISCUSSION

All scans were reconstructed directly on the scanner hardware in approximately 2 min/volume.
TA for different AF are reported in Figure 2 for T_1ρ-weighted images and maps. Images showed increased undersampling artifacts with increasing AF; as a result, the NRMSE increased and the SSIM decreased (Figure 2, bottom). Notably, similar visual image quality as well as NRMSE and SSIM were found for AF=4 and 5.1. Noise amplification in the center of the brain with generally poor image quality was found for AF≥7.1. Considering these results, AF=5.1 was selected as a good tradeoff between quality and TA, allowing for T_1ρ mapping in <7 min. With this protocol, the reconstruction regularization factor was empirically optimized and set to 0.001.
Representative orthogonal views from one subject using the optimized protocol illustrate the image quality obtained in T_1ρ-weighted images and maps (Figure 3).
In another subject, Repeat and One-Week scans show good image quality and qualitative similarity for all T_1ρ weightings as well as for the maps in comparison to their Reference scan (Figure 4).
Quantitatively, mean T_1ρ ranged between 95 and 110 ms in the VOIs with higher values for gray matter structures and thalamus (Figure 5A). Excellent agreement (R²=0.99) and repeatability (limits of agreement=1.2 ms) were found between Repeat and Reference measures (Figure 5B). Very good agreement (R²=0.96) and high reproducibility (limits of agreement=1.8 ms) were also found for One-Week measures (Figure 5C).

CONCLUSION

High-resolution T_1ρ mapping of the whole brain is feasible with the proposed method in under 7 min with images and maps directly obtained at the scanner with reconstruction times compatible with a clinical workflow. Excellent repeatability and good reproducibility of the technique were found in a small healthy cohort. In conclusion, the proposed method may ease the use of T_1ρ mapping for clinical research in both cross-sectional and longitudinal studies.

Acknowledgements

No acknowledgement found.