0668

In vivo visualization of white and gray matter sub-structures using fast quantitative T1 mapping of the human spinal cord at 7T with 300-µm in-plane resolution

Aurélien Massire^1,2,3, Henitsoa Rasoanandrianina^1,2,3, Manuel Taso^1,2,3,4, Arnaud Le Troter^1,3, Maxime Guye^1,3, Jean-Philippe Ranjeva^1,2,3, and Virginie Callot^1,2,3

¹Aix-Marseille Univ, CNRS, CRMBM, Marseille, France, ²iLab-Spine - Laboratoire international - Imagerie et Biomécanique du rachis, Marseille, France, ³APHM, Hôpital Universitaire Timone, CEMEREM, Marseille, France, ⁴Division of MRI Research, Department of Radiology, Beth Israel Deaconess Medical Center & Harvard Medical School, Boston, MA, United States

Synopsis

T₁ mapping of the cervical spinal cord at 7T with a 300-µm in-plane resolution using the MP2RAGE sequence is presented. A semi-automated post-processing pipeline enabled group studies (n=10 healthy subjects) with quantitative evaluation within small spinal cord sub-structures from C1 to C7, explored for the first time using T₁ mapping. This work lays the groundwork for improved characterization of spinal cord degeneration using 7T MRI, by allowing investigation within particular relevant regions such as the anterior/posterior gray matter horns and ventromedial white matter tracts. It also offers new perspectives to build dedicated spinal cord substructures templates.

Introduction

Magnetization-prepared rapid acquisition gradient echoes (MP2RAGE^1,2) is a fast and robust T₁ mapping sequence recently used^3,4 to quantitatively investigate the spinal cord (SC) both at 3T and 7T. The maximal isotropic spatial resolution reported so far⁴ at 7T for SC imaging was (0.7 mm)³. Although sufficient to globally describe tissue alteration, this resolution is insufficient to properly investigate SC sub-structures within the gray matter (GM) horns or the ventral white matter (WM) funiculi for instance.

In this work, we therefore propose to run MP2RAGE acquisitions at 7T with a non-isotropic resolution of 0.3x0.3x5 mm³ that minimizes in-plane partial volume effect (PVE) contamination hence enabling T₁ mapping of SC substructures along C1-to-C7 levels, including WM funiculi, GM motor columns and central canal, hence opening new perspectives for disease investigation and template building.

Methods

Experiments were performed on a whole-body 7T system (Siemens Healthcare) with an eight-channel transceiver cervical SC coil array (Rapid Biomedical), on ten healthy volunteers (29±13 years old [range 18-58]). T₁ mapping was performed with one or two 3D-MP2RAGE acquisitions, depending on the subject’s SC curvature. Ultra-high-resolution 2D anatomical imaging using a multi-echo GRE was used to reslice 3D T₁ volumes at each vertebral level (mid-vertebra slices) using c3D (ITK-Snap). To compensate for imperfect radiofrequency excitation biasing the T₁ estimation, an offline correction for B₁⁺-inhomogeneity⁴ was added using a B₁⁺ map acquired with a magnetization-prepared turbo-FLASH sequence⁵. Acquisition parameters of all sequences are provided in Figure_1.

Image post-processing was done using MATLAB (Mathworks) and FSL (FMRIB). T₁ maps were denoised using BM4D⁶. For each subject, the SC was semi-automatically segmented using PropSeg⁷. Images were then registered (affine (FLIRT) and non-linear (ANTs SyN⁸) registrations) to the oversampled AMU₄₀ T₂*-weighted template⁹ as an initial target. The mean T₁ map derived from the n=10 subjects averaging was then used as a target for subsequent iterative registrations (n=3) of all the individual T₁ data, in order to generate a sharper quantitative T₁ template.

As a preliminary step to define WM/GM substructures based on in vivo data, parcellation within ventral/lateral/dorsal WM fasciculi, anterior/posterior GM horns and central canal was performed manually based on prior histological knowledge¹⁰ and adjusted using T₁ thresholding. T₁ values within each defined region of interest (ROI) were then evaluated and statistical analyses were performed using multiple comparison Wilcoxon non-parametric tests on JMP9 (SAS).

Results

Examples of individual axial T₁ maps extracted from two MP2RAGE volumes aligned to the upper/lower cervical segments respectively are shown in subject space (Figure_2). Small structures such as the GM posterior horns, posterior septum and central canal could easily be delineated.

Figure_3 exhibits individual T₁ maps within the AMU₄₀ template space, mean resulting T₁ template and SC parcellation. Anatomical details and specific T₁ areas visualized on individual map could still be identified on the averaged maps.

For clarity sakes, only C2 and C4 quantitative results are reported. Mean and in-ROI standard deviation of T₁ values measured within the averaged T₁ map are reported for all considered ROIs in Figure_4. In addition, individual T₁ values and statistical results are shown in Figure_5. T₁ values were found significantly different for several ROIs (e.g. C2 ventral fasciculi and C4 medial GM horns different from their surroundings).

Discussion

The T₁ maps obtained in this work had unprecedented very high in-plane resolution, enabling exquisite anatomical details visualization and confident estimation of some so-far invisible SC-substructure mean T₁ values. T₁ quantification exhibited significant differences between several ROIs, which will likely be of great interest for further studies on SC pathologies. In-ROI and inter-subjects standard deviations were around 5% of the mean value.

Current limitations of this study include: 1/ vertebral levels instead of cervical levels¹¹, which should be taken into account to strengthen the SC structural subdivision, 2/ ROI delineation, which should be improved using objective methods such as clustering algorithms¹² and 3/ SC curvature, which should be positioned upright as much as possible since two non-isotropic MP2RAGE acquisitions were not always sufficient to adequately sample all vertebral levels with minimal PVE in the axial plane (9 slices over 70 discarded due to through-slice PVE contamination only).

Conclusion

Overall, this study opens new frontiers for an unprecedented in vivo characterization of the healthy and pathological SC substructures. These T₁ results, combined to additional contrasts (T₂* mapping and diffusion tensor imaging) will very likely provide accurate SC cyto/myelo-architecture atlases from C1 to C7 when used with clustering methods¹³. Additional work will be endeavored to exploit the whole 3D T₁ volumes to study the evolution along the cord of each individual substructure shape and mean T₁.

Acknowledgements

The authors would like to thank Véronique Gimenez and Lauriane Pini for the study logistics. This work was supported by the following funding sources: Investissements d’Avenir 7T-AMI-ANR-11-EQPX-0001, A*MIDEX-EI-13-07-130115-08.38-7T-AMISTART, A*MIDEXANR-11-IDEX-0001-02, Aix-Marseille Université, AP-HM, ARSEP and CNRS (Centre National de la Recherche Scientifique).

References

[1] Marques et al., NeuroImage 49, 2010. [2] Marques and Gruetter, PLoS One 8, 2013. [3] Rasoanandrianina et al., ISMRM 2017, p 912. [4] Massire et al., NeuroImage 143, 2016. [5] Fautz et al., ISMRM 2008, p 1247. [6] Maggioni et al., IEEE Trans Image Process 22, 2013. [7] De Leener et al., NeuroImage 98, 2014. [8] Avants et al., Medical Image Analysis 12, 2008. [9] Taso et al., NeuroImage 117, 2015. [10] Development of the human spinal cord, J Altman & SA Bayer, Oxford university press, 2001, p9, 46, 72. [11] Cadotte et al., AJNR 36, 2015. [12] Jain et al., Data Clustering: A Review. ACM Comput Surv 31, 1999. [13] Taso et al., ISMRM 2017, p 18. [14] Held et al., Journal of Neuroradiology 30, 2003.

Figures

Figure 1: Sequence parameters used in this study. In order to minimize partial volume effect from the subject cord curvature, a second MP2RAGE volume could be acquired using a tilted slab orientation. Additional specific parameters for the MP2RAGE sequence were: inversion times: 0.7/2.4s, parallel imaging acceleration factor: 2, and ‘LIN/PAR swap’ option enabled. GRE slices were positioned perpendicular to the cord at each cervical level and used to reslice 3D T₁ volume. B₁⁺ mapping was used to correct for biased T₁ estimation.

Figure 2: Imaging results on healthy volunteer. a: C1-to-C7 GRE slice positions are illustrated on a sagittal TSE image. b-c: MP2RAGE first (cyan) and second (purple) volume orientations are indicated on same TSE images. To minimize in-slice PVE contamination, C1-to-C4 slices are resampled on the first volume whereas C5-to-C7 slices are resampled on the second volume. d: High-resolution anatomical imaging using MEDIC¹⁴ reconstruction used to resample T₁ maps, from C1 to C7. e: Resulting B₁⁺-corrected, denoised, resampled to 0.18x0.18x2 mm³ resolution T₁ maps (in seconds), from C1 to C7, in the subject space, before registration.

Figure 3: Colored T₁ maps (in seconds) at C2 and C4 levels, registered in the oversampled AMU₄₀ reference space (spatial resolution: 0.18x0.18x2 mm³). a: individual T₁ map on one volunteer, b: averaged T₁ template (n=10) and c: corresponding manually segmented SC substructures. Manually segmented substructures are (in numerical order): ventral, inferior intermediate and superior lateral WM fasciculi, cuneate and gracile WM fasciculi, central septum, posterior GM horns, gelatinous substance, anterior GM (for C4: lateral, central and medial) and central canal. d: averaged T₁ maps at all levels (13 slices over 70 have been discarded due to PVE and artifacts).

Figure 4: Mean T₁ ± in-ROI standard deviation (in ms) measured for the first time in various SC substructures at C2 and C4 levels, in the average T₁ map (n=10 volunteers). Considered ROIs are: WM fasciculi (ventral, lateral inferior/intermediate/superior, gracile and cuneate), GM anterior horns (separated in three lateral, central and medial regions for C4 level), GM posterior horns, gelatinous substance, and cerebrospinal fluid-dominated regions (central canal and septum). For each ROI, the number of pixels is also provided (0.18x0.18x2 mm³ resampled resolution).

Figure 5: T₁ measurements for all ten healthy volunteers in all numbered SC substructures, at C2 and C4 levels. Individual points are displayed as black dots. The red box plot bars represent the minima, first quartile, median, third quartile and maxima. The green diamond centers represent group means while the extremities represent the 95% confidence interval. WM ROIs (light grey shade) and GM ROIs (grey shade) T₁ value differences were always statistically significant. Additional significant differences are displayed on the graphs (*p<0.05). Most interesting results reported are related to C2 ventral fasciculi and C4 medial GM horns.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)

0668