Synopsis

Keywords: Quantitative Imaging, Data Processing, Brain cortex

Quantitative MRI allows establishing normative atlases of relaxometry parameters which enable single-subject comparisons for anomaly detection. The large anatomical inter-subject variability of the brain cortex and its convoluted shape, however, complicate such comparison in this region. In this study, a method to align inter-subject brain cortices is proposed and a voxel-wise normative T₁ atlas in the cortex is built from a cohort of 133 healthy subjects scanned at 7T. The atlas is used to detect and characterize T₁ alterations in two multiple sclerosis patients on a single-subject basis.

Introduction

The potential of quantitative MRI (qMRI) to establish normative atlases that enable single-subject comparisons to detect subtle brain tissue abnormalities has been recently shown for white matter (WM)^1-5. The extension of this method to cortical gray matter (GM) would allow a more comprehensive characterization of brain diseases that affect this region. However, given the large inter-subject anatomical variability of the cortex and its highly convoluted shape, mapping cortical locations across subjects is challenging. Using spherical representation of the cortex⁶, a cortical T₁ atlas was recently established at 3T⁷. Generally, high resolution (<$$$\,$$$1$$$\,$$$mm³) images are beneficial to properly characterize cortical GM due to its thin, layered structure.
In this study, a normative atlas of cortical T₁ values at 7T is established from MP2RAGE scans^8,9. As a proof of concept, the atlas is used to detect pathology-induced T₁ alterations in two multiple sclerosis patients.

Methods

Population and MR acquisition
A healthy cohort of 133 subjects (74 females, median age$$$\,$$$=$$$\,$$$28$$$\,$$$y/o, range$$$\,$$$=$$$\,$$$[15-74]$$$\,$$$y/o) and two patients (29 and 37$$$\,$$$y/o, males) with relapsing-remitting multiple sclerosis (RRMS) were scanned at 7T (MAGNETOM Terra, Siemens Healthcare, Erlangen, Germany). T₁-weighted images and T₁ maps were acquired simultaneously using a MP2RAGE research application sequence^8,9 (resolution$$$\,$$$=$$$\,$$$0.6$$$\times$$$0.6$$$\times$$$0.6$$$\,$$$mm³, FOV$$$\,$$$=$$$\,$$$240$$$\times$$$240$$$\times$$$172$$$\,$$$mm³, TI₁/TI₂$$$\,$$$=$$$\,$$$800$$$\,$$$ms$$$\,$$$/$$$\,$$$2700$$$\,$$$ms, TR$$$\,$$$=$$$\,$$$6s, undersampling:$$$\,$$$CSx4, TA$$$\,$$$=$$$\,$$$7:49$$$\,$$$min). All scans were performed using a 1-channel TX/32-channel RX head coil (Nova Medical, Wilmington, MA).

Brain cortical alignment
The total intracranial volume (TIV) was extracted using the MorphoBox research application on the MP2RAGE T₁-weighted uniform (“UNI”) contrast^10,11. The resulting TIV mask was used to strip the skull in T₁ maps.
The FreeSurfer segmentation and brain surface reconstruction pipeline¹² was used to align the brain cortices of each healthy subject to subsequently build a voxel-wise normative T₁ atlas. To that end, the skull-stripped UNI volumes were registered to the 1$$$\,$$$mm isotropic FreeSurfer MNI305 template (“fsaverage”) with a surface-based method⁶, and the same transformation was applied to the skull-stripped T₁ maps. The interfaces between GM and cortical cerebrospinal fluid (“pial surface”) and between WM and GM (“white surface”) were extracted as surfaces in the fsaverage space. To extract T₁ values in the whole cortical GM, the white surface was projected along its normal vectors towards the pial surface using specific distances to reach different cortical depths. T₁ values were extracted at six equally spaced cortical depth fractions, between and including the white (0.0) and pial (1.0) surfaces, and four equidistant juxtacortical layers (hereafter reported as negative cortical depth fractions), obtained by projecting the white surface in the opposite direction.

Normative atlas and single-subject comparison
A voxel-wise normative atlas for cortical and juxtacortical T₁ values was computed by modeling the expected T₁ in the fsaverage space as:$$E\{T_1\}=\beta_0+\beta_{sex}\cdot\,sex+\beta_{age}\cdot\,age+\beta_{age^2}\cdot\,age^2\,\textrm{,}\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,(1)$$where $$$\beta_0$$$ is the model intercept, and sex$$$\,$$$=$$$\,$$$1 if the subject is male, 0 if female. Age was centered at the average age of the healthy cohort (33$$$\,$$$y/o), and $$$\beta_{age^2}$$$ was introduced to consider the quadratic evolution of T₁ with age¹³.
To detect abnormal T₁ values on a single-subject basis, deviation from normative ranges of measured T₁ was assessed at each voxel by z-scores:$$z_{T_1}=(T_1-E\{T_1\})/RMSE\,\textrm{,}\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,(2)$$with the root mean square error (RMSE) being an estimation of the standard deviation of the residual errors across the linear model of Eq. 1. To evaluate the robustness of the proposed method with respect to false positives, T₁ z-scores were calculated within the healthy cohort with a 10-fold cross-validation to yield the false positive rate (FPR) for each voxel by considering |z_T1|$$$\,$$$>$$$\,$$$2 as abnormal³. Additionally, to demonstrate the potential of the proposed framework for automatic detection of potentially abnormal tissue, z-scores were computed for two patient datasets at each cortical depth, resampled to the patients’ native space and concatenated into a deviation map.

Results

Representative slices of the established cortical T₁ atlas are shown in Figure$$$\,$$$1. T₁ values were found to decrease from the pial surface to the juxtacortical WM, in accordance with previous findings¹⁴ (Figure$$$\,$$$2). As an example, average T₁ values at cortical depth fractions 0.4 and -0.4 were found to be E{T₁}$$$\,$$$±$$$\,$$$RMSE$$$\,$$$=$$$\,$$$1865$$$\,$$$±$$$\,$$$131$$$\,$$$ms, and 1330$$$\,$$$±$$$\,$$$108$$$\,$$$ms, respectively. Highest variability (median COV$$$\,$$$>$$$\,$$$10%) was found at the atlas boundaries, i.e., at cortical depth fractions 1.0 and -0.8 (Figures 1$$$\,$$$-$$$\,$$$2).
FPR was found to be comparable across cortical depths (Figure$$$\,$$$3). Overall FPR values were: median$$$\,$$$=$$$\,$$$2.25$$$\,$$$%, range$$$\,$$$=$$$\,$$$[0-24.81]$$$\,$$$%.
The deviation map of patient #1 shows both juxtacortical lesions (Figure$$$\,$$$4A:$$$\,$$$highest z_T1$$$\,$$$=$$$\,$$$11.88, Figure$$$\,$$$4B:$$$\,$$$highest z_T1$$$\,$$$=$$$\,$$$5.44) and diffuse alterations of the normal-appearing GM (NAGM) tissue (Figure$$$\,$$$4B). Another example of a juxtacortical lesion (highest z_T1$$$\,$$$=$$$\,$$$12.38) is shown in the deviation map of patient #2 (Figure$$$\,$$$5).

Discussion and Conclusion

In this study, a normative T₁ atlas for cortical GM and juxtacortical WM at 7T was created. It proved to be robust with overall low FPR. Its potential for identifying brain tissue abnormalities was illustrated with data from two RRMS patients, where the proposed method detected both focal lesions and diffuse T₁ changes in the NAGM. Future work should focus on further validating the method in a larger cohort of patients, ideally including also different pathologies affecting the cortex. Additional efforts should also concentrate on developing an atlas while preserving the native high-resolution.
In summary, these encouraging results may enable accurate and sensitive disease characterization in cortical GM and juxtacortical WM at 7T on a single-subject basis.

Acknowledgements

No acknowledgement found.