Introduction

Recent studies on brain relaxometry have shown the potential of comparing single-patient data to atlases of normative tissue parameters for the personalized characterization of pathological tissue in white matter (WM)^1–4. However, achieving the same goal in cortical gray matter (GM) is challenging; the complex cortical folding patterns as well as the large anatomical inter-subject variability of the cortices hinder an accurate spatial normalization of MR data.
To address this issue, this study introduces a method for inter-subject brain cortex alignment to enable the comparison of T₁ measurements in cortical GM between an atlas of normative relaxation times and a map acquired in a single subject. The feasibility and sensitivity of the established atlas in the detection of tissue alterations is demonstrated in three case reports.

Methods

Study population and MR protocol
Two cohorts of healthy individuals were scanned in two different centers:

• Site A: 201 subjects (123 females, age = [20-64] y/o);
• Site B: 84 subjects (53 females, age = [21-58] y/o).

T₁ mapping was achieved in both centers with the MP2RAGE sequence⁵ using the protocol parameters detailed in Table 1. Both cohorts were scanned at 3T (MAGNETOM Skyra and Prisma, Siemens Healthcare, Erlangen, Germany) using a 64-channel head/neck coil. Written informed consent was obtained prior to examination.
For a proof-of-concept, MP2RAGE data were also acquired in a patient (female, 36 y/o) with focal cortical dysplasia (FCD) (Table 1, protocol A) and two patients (male, 53 y/o; and female, 61 y/o) with multiple sclerosis (MS) (Table 1, protocol C), in agreement with the institutional regulations. One radiology resident and one neuroradiologist (4 and 20 years of experience, respectively) manually segmented abnormal tissue regions in the FCD data.
Brain cortices alignment
To enable the inter-subject and voxel-wise comparison of T₁ relaxation times in cortical tissues, the brain surface reconstruction pipeline of FreeSurfer was employed⁶. A watershed algorithm was first used to remove the skull from the MP2RAGE uniform T₁-weighted contrast (“UNI”)⁷. After spatially registering the skull-stripped UNI image onto the FreeSurfer MNI305 template, the exact spatial locations of the WM/GM boundaries (“white surface”) and GM to cortical cerebrospinal fluid boundaries (“pial surface”) were extracted with a surface-based algorithm⁸. The healthy subjects’ cortices were subsequently resampled onto the FreeSurfer average surfaces, resulting in accurate brain alignment that accounts for inter-individual differences in cortical folding patterns. To extract T₁ values from multiple cortical layers, vectors orthogonal to the white surface at each vertex along the surface were computed and used to project the white surface along them. A projection fraction of 1/5 was used to obtain four GM layers, at 20%, 40%, 60% and 80% depth of the cortex (with 0% and 100% depths corresponding to the white and pial surfaces, respectively). Similarly, four juxtacortical WM layers were estimated by projecting the white surface in the inward direction (see Figure 1).
Normative atlases
A mixed-effects model (fixed effects: sex, age, age²; random effect: site) was used to establish a voxel-wise normative atlas of T₁ relaxation times in the FreeSurfer average space:$$\mathit{E}\left\{T_{1,j}\right\}=\ {(\beta}_0+u_{0,j})+\beta_{sex}\ast sex+\beta_{age}\ast age+\beta_{{age}^2}\ast{age}^2\qquad\qquad \textrm{(Eq. 1)}$$with $$$j$$$=(1,2) referring to the two sites, $$$\beta_i$$$ being the coefficients of the fixed effects and $$$u_{0,j}$$$ the coefficient of the random effect.
Method for single-subject comparison
To characterize abnormal T₁ values on a single-subject basis, the T₁ map of a patient is transformed into the FreeSurfer average space by applying the same processing as previously described. Abnormal T₁ values are then characterized in a voxel-wise manner by computing z-scores within each surface. Subsequently, each resulting surface in the FreeSurfer average space is resampled, together with its z-scores, into the corresponding subject’s native surface, converted back to volumetric intensities, and concatenated into a unique deviation map.

Results

Example axial slices of the established T₁ atlas are shown in Figure 2. The average T₁ values (i.e., intercept coefficient at mean age of the healthy cohort) were found to range from 745 ms to 2187 ms according to the cortical depth and to differ among brain regions. The RMSE of the models was found to increase when moving from WM to GM and to be higher in the frontal lobe, a brain region that also showed the largest T₁ variations due to the different sites (Figure 2).
The deviation map computed from the FCD patient’s data identified an area of abnormally reduced T₁ values (mean z-score of -2.25), which corresponded with the expert’s manual segmentation (Figure 3). Lesions in juxtacortical WM and cortical GM of the MS patients were also detected by the method (Figure 4), exhibiting increased T₁ values (i.e., positive z-scores).

Discussion and Conclusion

This work introduced a new method to determine normative T₁ values in the juxtacortical WM and cortical GM. The potential to detect cortical tissue alterations with the established atlas was demonstrated in three case patients with neurological conditions affecting cortical regions. The resolution of the acquired images limits the maximum number of cortical layers that can be extracted. Ultra-high field strength imaging thus represents a good option for future work to overcome this limitation.
In conclusion, promising proof-of-concept results were obtained which could enable robust pathology characterization on a single-subject basis in cortical GM.

Acknowledgements

No acknowledgement found.