New rapid, accurate T2 quantification detects pathology in normal-appearing brain regions of relapsing-remitting MS patients.

INTRODUCTION: Quantitative T2 mapping may provide an objective biomarker for occult nervous tissue pathology in relapsing-remitting multiple sclerosis (RRMS). We applied a novel echo modulation curve (EMC) algorithm to identify T2 changes in normal-appearing brain regions of subjects with RRMS (N = 27) compared to age-matched controls (N = 38).
METHODS: The EMC algorithm uses Bloch simulations to model T2 decay curves in multi-spin-echo MRI sequences, independent of scanner, and scan-settings. T2 values were extracted from normal-appearing white and gray matter brain regions using both expert manual regions-of-interest and user-independent FreeSurfer segmentation.
RESULTS: Compared to conventional exponential T2 modeling, EMC fitting provided more accurate estimations of T2 with less variance across scans, MRI systems, and healthy individuals. Thalamic T2 was increased 8.5% in RRMS subjects (p < 0.001) and could be used to discriminate RRMS from healthy controls well (AUC = 0.913). Manual segmentation detected both statistically significant increases (corpus callosum & temporal stem) and decreases (posterior limb internal capsule) in T2 associated with RRMS diagnosis (all p < 0.05). In healthy controls, we also observed statistically significant T2 differences for different white and gray matter structures.
CONCLUSIONS: The EMC algorithm precisely characterizes T2 values, and is able to detect subtle T2 changes in normal-appearing brain regions of RRMS patients. These presumably capture both axon and myelin changes from inflammation and neurodegeneration. Further, T2 variations between different brain regions of healthy controls may correlate with distinct nervous tissue environments that differ from one another at a mesoscopic length-scale.

Introduction

Relapsing-remitting multiple sclerosis (RRMS) is a common neurological disease affecting young adults and characterized by recurrent clinically-symptomatic episodes of inflammation and the insidious progression of disability. The classic MRI hallmarks for RRMS are transient foci of contrast enhancement during acute inflammatory episodes and the gradual accumulation of T2 or FLAIR hyperintensities. Previous studies have established that foci of T2 prolongation correlate with inflammation, edema, demyelination, abnormal re-myelination, gliosis and/or axonal loss (Laule et al., 2011, Laule et al., 2013, Lund et al., 2012, MacKay et al., 2006). Although not all T2-bright lesions are MS-related (Liu et al., 2013), the number and locations of focal T2 hyperintensities can often help support the clinical diagnosis of MS (Polman et al., 2010). However, T2 changes from subtle tissue pathology can be hard to detect visually on clinical MRI scans particularly during early stages of the disease. Further, visually-apparent T2 lesions correlate poorly with patient disability or MS disease progression (Barkhof, 2002).

This radiology-pathology discordance in MS patients has been attributed to inflammation and neurodegeneration that remain occult to visible detection on conventional MRI. Several advanced techniques, such as MR spectroscopy, diffusion, and magnetization transfer MRI, have demonstrated abnormalities in normal-appearing brain regions for MS patients (Ceccarelli et al., 2007, Davie et al., 1997, Mangia et al., 2014). Subtle T2 differences have also been observed in normal-appearing white matter as well (Bonnier et al., 2014, Laule et al., 2004). While each of these techniques may have clinical value for the early diagnosis of MS and monitoring disease progression, quantitative methods also may provide more objective markers of specific pathological components of MS. However, to realize this potential, these methods need to be accurate, stable and reproducible across different MRI protocols and imaging sites – features that remain elusive for diffusion, magnetization transfer and conventional T2 mapping techniques.

Accurate T2 mapping in clinically-feasible scan times is challenging due to the inherent bias of rapid multi spin-echo (MSE) sequences by stimulated and indirect echoes, non-rectangular slice profiles and transmit-field (B1+) inhomogeneities. Furthermore, this bias depends on the pulse sequence implementation and scan parameters, causing T2 values in the same subject to vary between scanners and protocols (Deoni et al., 2003, Lebel and Wilman, 2010, Poon and Henkelman, 1992). T2 estimation accuracy may be improved with short TR single spin echo (SSE) (Sussman et al., 2010), analytical solutions to coherence pathways in MSE acquisitions (Lebel and Wilman, 2010, Lukzen et al., 2009, Prasloski et al., 2012), model-based reconstruction approaches (Huang et al., 2012), modeling signal with non-spin echo based pulse sequences (Deoni et al., 2003, Schmitt et al., 2004, Warntjes et al., 2007), or heuristic signal model approaches (Ma et al., 2013). We have recently reported an alternative approach for accurate quantitative T2 mapping, the echo modulation curve (EMC) algorithm, which uses a Bloch simulation to trace all coherence pathways, including all stimulated and indirect echoes during an MSE acquisition. This algorithm accounts for different slice profiles, radiofrequency pulse shapes, crusher gradients, and spin relaxation during the radiofrequency (RF) pulses (Ben-Eliezer et al., 2015a). Because the EMC algorithm incorporates the specific implementation of the MSE pulse-sequence, this method is robust to different acquisition strategies, and offers reliable mapping in clinically feasible scan times. Note that acquisitions for EMC fitting can be tailored to emphasize speed and/or accuracy for particular T2 components. Further info can be found in a recent report including detailed analysis of T2 mapping approaches, and demonstrating the advantage of Bloch-simulation-based approach, like EMC, over extended phase-graph (EPG) techniques (McPhee and Wilman, 2016).

Here, we used the EMC algorithm to characterize different brain regions in a cohort of clinical RRMS patients compared to healthy age-matched controls. Using this technique we were able to detect subtle, yet statistically significant, anatomy-specific T2 differences within normal-appearing gray and white matter structures in RRMS patients. We also observed interesting T2 differences in healthy control subjects between individual brain structures with different anatomic locations or specific functions.

Materials & methods

Subject enrollment & MRI protocol

This study was performed with approval from the local institutional review board. The MSE sequence was part of the routine noncontrast head protocol for patients with an established clinical diagnosis of multiple sclerosis, referred from our academic center's MS neurology specialists and scheduled on an outpatient 3-T MRI scanner with a 20-channel head & neck coil (Skyra or Prisma, Siemens Healthcare, Erlangen, Germany). MSE scan parameters were: TR = 2500 ms, Echo-spacing = 12 ms, First echo-time = 12 ms, Nechoes = 10, res = 1.7 × 1.7 mm2, Nslices = 23, slice-thickness = 3 mm, bandwidth = 200 [Hz/Px], Tacquisition = 2:44 min using 2 × GRAPPA acceleration. The use of longer echo trains could theoretically improve T2 fitting accuracy (Whittall et al., 1997), particularly if long T2 components contribute to voxel signal (e.g. in more cystic MS lesions). These are not the focus of the current work and investigation of such tissues is left for future study. However, the specific absorption rate (SAR) and scan-time limitations when scanning patients in clinical settings, limited the number of echoes that could be used while keeping sufficient volumetric coverage (i.e., number of sliced). Also note that since intravenous contrast was not ordered in these subjects, the ordering clinician's suspicion for active inflammatory lesions was low. The standard MRI protocol also included sagittal 3D SPACE FLAIR, axial susceptibility-weighted imaging, and a 3D 1-mm isotropic volumetric MPRAGE sequence. A board-certified neuroradiologist confirmed typical MRI findings consistent with clinical MS (Polman et al., 2010). A board-certified neurologist reviewed the electronic medical record to confirm MS diagnosis and subtype, disease duration and the most recent documented patient-reported expanded disability scale score (PDSS) (Hohol et al., 1995). PDSS is similar, and correlates strongly, with the expanded disability status scale (EDSS) (Learmonth et al., 2013). PDSS, however, is more efficient to collect in routine clinical care. A PDSS score of “1” indicates mild disability, “2” indicates moderate disability without gait impairment, “3” reflects gait impairment without use of a cane, and “4–5” indicate early use of a cane (after 25 ft of walking) vs late use of cane (required to walk even 25 ft). During this chart review, 2 subjects with secondary progressive MS and 1 subject with primary progressive MS were identified and excluded from the study. Overall, 27 subjects with RRMS (19 females, mean age 48.5 ± 9.2 years/o) with mean disease duration of 12.6 ± 8.6 years and PDSS of 2 ± 1.8 (no units) were included in this study. A PDSS of 2 corresponds to significant problems related to MS such as visual impairment, sensory symptoms, or fatigue, but no limitations in walking ability. Age-matched control subjects (N = 38, 15 females, mean age 39.0 ± 9.6 years/o) without history of neurological disease or known white matter hyperintensities were recruited from the local community.

T2 relaxation mapping

T2 maps were generated via (a) pixel-by-pixel fitting of the MSE time series of DICOM images to a theoretical exponential decay of the form of S(t) = S0 ∙ e− (Levitt, 2001) and (b) using the EMC algorithm (Ben-Eliezer et al., 2015a). All fitting procedures were programmed in-house using C ++ and MATLAB (The MathWorks Inc., Natick, MA). The EMC algorithm consists of an initial pre-processing stage, in which Bloch simulations of the prospective MSE protocol are performed using the exact RF pulse shapes and other parameter values used on the MRI scanner. This allows EMC to replicate the actual decay curve in MSE protocols and produce T2 values that are independent of the particular choice of experimental parameter set. Simulations are repeated for a range of T2 relaxation values and transmit-field (B1+) inhomogeneity levels (T2 = 1 … 1000 ms, B1+ = 70 … 130%), producing a database of decay curves each associated with a unique [B1+, T2] value pair. Once experimental MSE data are acquired, the signal time series from each pixel is matched to the simulated database of EMCs by calculating the L2 norm of the difference between the experimental and simulated decay curves, and choosing the database entry that yields the minimum norm. This minimization procedure is implemented using a full search over the entire database, which is completed in ~ 15 s per slice. A unique T2 value associated with the matched EMC is then assigned to the corresponding pixel, eventually yielding the desired parametric map after the procedure is repeated for all pixels in the slice. To avoid fitting bias due to Rician noise, signal decay curves are truncated below 10% of the first time-point intensity – for both exponential and EMC fitting. This resulted in exclusion of 1–2 echoes in areas of very short T2 values. Lastly, proton density (PD) maps are calculated by extrapolating the image from the first echo time to time t = 0 s based on the calculated T2 map under the assumption that purely exponential decay takes place between spin excitation and the first acquisition event.

Region of Interest (ROI) analysis

Volumetric segmentation of six ROIs was performed using automated FreeSurfer software (http://surfer.nmr.mgh.harvard.edu/) based on the three-dimensional T1-weighted MPRAGE data. The ROIs included: global white matter, cortical gray matter, thalamus, caudate nucleus, putamen and globus pallidus. ROIs were then delineated on the T2 maps by co-registering these maps to FreeSurfer atlas space using FreeSurfer conversion and registration tools. Lastly, the perimeter of these automated ROIs was eroded by one pixel in order to avoid edge artifact such as partial volume effects.

Manual ROIs were drawn by a board-certified neuroradiologist on 5 different axial T2-weighted images (A-E) with TE = 81 ms, similar to standard clinical T2-weighting (Fig. 1). At slice A, 6 ROIs were drawn for the bilateral optic radiations, temporal stem white matter and globus pallidus. At slice B, 8 ROIs were drawn for the bilateral caudate nucleus, putamen, thalamus and posterior limb internal capsule. At slice C, 2 ROIs were drawn for the genu and splenium of the corpus callosum. At slice D, a single ROI was drawn for the body of the corpus callosum. At slice E, 2 ROIs were drawn for the bilateral centrum semiovale. As an internal control, additional smaller ROIs were drawn within T2-bright lesions located in the periventricular, subcortical and juxtacortical white matter for each of the RRMS subjects and compared to similarly located ROIs in healthy controls (see Fig. 2). For all other ROIs, T2-bright lesions were avoided. In some cases this meant no data could be obtained for specific ROIs in individual RRMS subjects.

Fig. 1

Expert-drawn manual regions-of-interest for a single subject with relapsing-remitting multiple sclerosis. 5 axial slices were included for the analysis; immediately below the anterior commissure (a), at the foramen of Monro (b), 1 cm above the foramen of Monro (c), roof of the lateral ventricles (d) and 1 cm above the lateral ventricles (e).

Fig. 2

Examples of regions-of-interest for MS lesions including (a) periventricular, (b) juxtacortical and (c) subcortical white matter. ROIs were drawn on a T2 weighted image (TE = 81 ms) synthesized based on EMC fitted T2 map without interpolation (note, black voxels indicate non-valid fitting results). Similar ROIs were drawn for investigating normal appearing brain regions, i.e., where lesions were not visibly present to a board-certified neuroradiologist (see Fig. 1).

Analysis of covariance (ANCOVA) was used to compare controls and RRMS subjects for the mean T2 value in each ROI adjusted for both age and sex. The error variance was allowed to differ across comparison groups to remove the assumption of variance homogeneity. Logistic regression and area under the ROC curve (AUC) was used to assess the utility of each measure to discriminate patients with and without lesions. Pearson correlation coefficients were used to assess the relationship between ROI T2 values and disease duration (years) and the ordinal PDSS. All statistical tests were conducted with SPSS 22 (IBM, Armonk, NY) at the two-sided 5% significance level. The purpose of this study was to identify potential T2 differences from many different brain regions so corrections for multiple comparisons was not conducted a priori – this may increase the risk of Type 1 errors but reduces the chance of detecting small T2 differences for investigation for future studies.

Results

Exponential and EMC T2 fitting comparison

Table 1 presents comparison of exponential and EMC fitting for 5 representative brain regions, and across all 38 healthy control subjects. Values demonstrate an average increase of 59% in mean T2 and 108% larger standard deviation for exponential fitting compared to EMC fitting (p < 0.001, all 5 regions). The EMC T2 values are very similar to previously published T2 values for these regions, when compared against a reference single spin echo (SSE) based T2 values (Ben-Eliezer et al., 2015a, Ben-Eliezer et al., 2016). SSE scans, however, require 50 min or more per scan and would not be practical in this larger study, especially for RRMS patients. Given these findings, we focus solely on EMC analysis for the remainder of this report.

Table 1

Mean, standard deviation (SD) and coefficient of variation (CV) of T2 values in five brain regions for a single healthy volunteer. Values are calculated across the set of 24 scans used to test stability of EMC vs. Exponential fitting methods (see text for additional info). In addition to providing the correct T2 value based on single spin echo reference (Ben-Eliezer et al., 2015a, Ben-Eliezer et al., 2015b), the EMC fitting algorithm produced lower SD and CV across all brain regions [WM: white matter; CC: corpus callosum].

ROI name	T2 values: Mean ± SD [ms] (CV %)
	EMC Fitting	Exponential fitting
Genu of CC	55.9 ± 2.1 (3.8)	87.6 ± 4.4 (5.0)
Splenium of CC	58.6 ± 2.7 (4.5)	93.8 ± 5.6 (6.0)
Caudate nucleus	56.0 ± 2.2 (4.0)	86.8 ± 5.5 (6.4)
Juxtacortical WM	53.0 ± 1.5 (2.9)	87.6 ± 3.7 (4.3)
Periventricular WM	63.5 ± 2.8 (4.3)	100.9 ± 3.6 (3.6)

Baseline and RRMS-specific differences in brain T2

In the FreeSurfer ROI analysis (Fig. 3) statistically significant increases were observed for most deep gray matter structures, with the largest observed difference being an 8.5% increase in thalamus T2 for RRMS subjects. Age did not affect thalamic T2 in control subjects and showed only a very weak positive correlation in RRMS subjects (R = 0.0154, p < 0.0001) (Fig. 4) . An ROI of global white matter that excluded regions of T1 hypointensity did not detect significant T2 differences between the groups. Conversely, the manual ROI analysis demonstrated several normal-appearing white matter regions with either statistically increased (temporal stem, body and genu corpus callosum) or decreased T2 values (posterior limb internal capsule) (Table 2). Potential explanations for the differences between FreeSurfer and manual ROI analyses are discussed below. It should also be noted that for all normal-appearing brain regions the intra-ROI SD – which reflects T2 (and general) tissue heterogeneity – was consistently higher in the RRMS patients, for both FreeSurfer and manual segmentations. Manual ROIs purposefully drawn in juxtacortical, subcortical and periventricular white matter lesions (obvious to a neuroradiologist on the T2-weighted reconstructed images) demonstrated 35.5–39.1% T2 increases compared to homologous regions in healthy controls (all comparisons, p < 0.001). We also observed statistically significant T2 differences between multiple gray and white matter regions in healthy controls (Table 2 and Fig. 5). Analysis of the manually segmented ROI demonstrated no significant left-right T2 differences for controls or RRMS patients.

Fig. 3

(a) Bar graph comparing T2 values (mean ± SD) in brain regions segmented using Freesurfer, in relapsing-remitting multiple sclerosis patients and in age-matched healthy controls (N = 27 & 38 subjects respectively). Regions that are statistically different between the two groups (p < 0.05) are denoted with “*”. (b) Representative FreeSurfer segmentation map, overlaid on an axial T1-weighted image at the level of the internal capsule.

Fig. 4

Mean thalamic T2 values using the manual segmentation were 8.5% higher in subjects with RRMS compared to age-matched healthy controls (p < 0.001). In controls, thalamic T2 values were independent of subject age (a). T2 values in RRMS patients showed minimal dependence on age (b).

Table 2

Summary of T2 values from healthy controls and patients with relapsing-remitting multiple sclerosis (RR-MS) in different regions of the brain, manually segmented by expert neuroradiologist (mean ± SD). Statistically significant differences (p < 0.05) are highlighted in green.

Fig. 5

Manual region-of-interest T2 values (mean ± SD, N = 38) for periventricular (PV), subcortical (SC) and juxtacortical (JC) white matter (WM) regions in healthy controls decreased in a centrifugal manner (all three groups statistically different, p < 0.05). This trend may reflect different underlying mesoscopic tissue environments such as variations in axon caliber, dispersion, packing density, myelination or even orientation with respect to the MRI main magnetic field.

Potential diagnostic utility of EMC T2 maps

We also tested the hypothesis that T2 values in brain regions without T2-bright lesions might be used to discriminate between controls and RRMS subjects. The FreeSurfer-segmented whole-thalamus ROI was found to be an excellent discriminator between the two groups (AUC = 0.913) – this accuracy left insufficient room for improvement by the addition of a second regional T2 measure to the diagnostic model. Another secondary aim of this retrospective study was to determine whether T2 values in individual brain regions, correlated with readily available clinical information in the electronic medical record for RRMS subjects. There was a modest negative correlation between RRMS disease duration, and the T2 values in normal-appearing juxtacortical white matter (r = − 0.46, p = 0.017). We also observed statistical trends (p < 0.1) for correlations between RRMS disease duration and subcortical white matter lesion and thalamic T2 increases. We did not observe regional T2 correlations to patient-reported disability scale score (PDSS).

Discussion

T2 changes in RRMS

Automated segmentation using FreeSurfer detected a statistically significant 8.5% increase in the thalamus T2 (p < 0.001). The thalamus can be affected early in the natural history of MS as a result of demyelination, iron accumulation, and axonal and synaptic loss (Gilmore et al., 2009). The thalamus is a major relay station within the nervous system such that the observed MS-related thalamic T2 changes may also reflect sequelae of retrograde degeneration from its widespread cortical and subcortical connections. Thalamic neuropathological changes and volume loss correlate with clinical symptoms of fatigue, dystonia, memory impairment and decreased processing speed (Minagar et al., 2013). Additional statistically significant T2 increases were observed with automated analysis for other gray matter structures, including cortex, pallidum and putamen (Fig. 2). Previous studies reported decreased T2-weighted image intensity (and hence inferred shorter T2 and T2⁎) in the caudate nucleus, putamen, globus pallidus and thalamus for clinically-isolated syndrome and MS patients. These differences progressed over time in individual subjects and were attributed to microscopic iron accumulation (Khalil et al., 2009). The current results may differ since, compared to signal intensity variations on T2-weighted images in previous reports, the multi spin-echo data used by the EMC algorithm is less sensitive to diffusion-mediated T2⁎ signal attenuation.

Manual ROIs within gray matter regions did not detect statistical differences between RRMS subjects and healthy controls. There are 3 potential explanations for this discordance. First, the automated method holds higher statistical power by including the entire structure volume, whereas manual segmentation delineated 2D ROIs in specific axial slices (due to the time-intensive nature of such segmentation). Alternately, the automated segmentation could have included lesions, which were excluded by the expert reader during manual segmentation. Finally the manual segmentation was performed on single axial slices, which may have inadvertently excluded a region of the thalamus more affected by RRMS pathology. The quantitative pathology literature does not make such a distinction with the data, suggesting RRMS pathologic changes in the thalamus are randomly distributed although there are anecdotal comments that the anterior and medial nuclear groups may be more affected (Vercellino et al., 2009). A future study could prospectively attempt detailed, blinded, time-intensive manual thalamic parcellation to determine how specific thalamic nuclei are affected, and whether such T2 changes correlate with clinical symptoms of fatigue, dystonia, memory impairment and reduced processing speed.

An advantage of the manual segmentation approach is the ability to sample specific white matter sub-regions, while automated FreeSurfer segmentation may include T1 isointense, yet T2 bright lesions, thereby confounding localized T2 changes in normal-appearing regions. Hence, with this approach, EMC based T2 values reflected subtle tissue changes for a variety of normal-appearing white matter structures in RRMS patients compared to healthy controls. Using manual regions-of-interest segmentations that excluded visible lesions on conventional T2-weighted images, we observed small but significant increases in T2 values for the corpus callosum in RRMS subjects (e.g. 11.4% increase in the body of the callosum, p = 0.012) (Table 2). In addition to changes in the T2 value, we also observed consistent increases in T2 variance for different brain regions in RRMS subjects compared to controls, suggesting higher heterogeneity in the underlying RRMS pathology burden. As expected, visible lesions on conventional MRI had T2 values 35–39% higher than homologous white matter regions in healthy controls (all comparisons, p < 0.001) – this supports the external validity of this T2 measurement technique, but was not the focus of the current study.

T2 hyperintensity in obvious RRMS lesions can result from multiple concurrent changes in the local mesoscopic tissue environment, including inflammation, edema, blood-brain-barrier breakdown, abnormal re-myelination, gliosis or axonal loss – it is difficult to attribute T2 changes to a single specific microstructural pathologic change (Lund et al., 2012, MacKay et al., 2006). Besides these pathological changes, additional biophysical processes may also contribute to T2 changes observed for normal-appearing brain tissue, including oligodendrocyte apoptosis with microglial activation (Barnett and Prineas, 2004), altered neuropil (Moll et al., 2008), dysregulated iron deposition (Stankiewicz et al., 2007) and/or changes to myelination and reduced axonal diameter from prior tissue repair (Dula et al., 2010). The relative contributions of these pathologies may vary between individuals, anatomic regions and at different times during the natural history of RRMS. These results may justify future longitudinal studies in patients and radiology-pathology quantitative T2 study in postmortem MS brains.

Several recent reports focused on multi-compartment T2 fitting schemes to increase the pathologic specificity of T2 relaxation measurements by correlating tissue histological changes to T2-based estimations of myelin water fraction (MWF) (MacKay et al., 2006). Accordingly, MWF appears reduced 16% and water content increased 2% in normal-appearing white matter (Laule et al., 2004). However, MWF is challenging to estimate and lacks a gold-standard method that will provide stable and reliable values. Some potential confounds to calculating MWF include dependence on intercompartmental water exchange and on acquisition strategies (Dula et al., 2010, Harkins et al., 2012, Zhang et al., 2015a), or the failure to distinguish between intact myelin and myelin debris (Alonso-Ortiz et al., 2015). Magnetization transfer might also affect MWF calculation, although recent data suggests that this factor may not influence MSE-based MWF measurements (Zhang et al., 2015b). Our current data has not been subjected to this type of analysis, but it is our perspective that accurate T2 estimation in RRMS patients with the EMC approach can eventually contribute to improved pathological specificity in particular brain regions or diseases. A limitation of the current EMC-derived single component T2 fitting is that microstructural tissue pathologies that affect individual T2 compartments independently, may be obscured or mitigate one another in a globally averaged measurement. The range of echo times used for data acquisition may also affect EMC estimation accuracy if very short and long T2 components are intermixed in the same voxel.

Anatomic variation of T2 in the healthy brain

Significant T2 differences were observed between normal brain regions in the healthy control group. Mean T2 values in the range of 70…90 ms for different brain regions (29% variance) beyond simple gray-white matter distinctions have been reported previously using exponential fits (Whittall et al., 1997). More recently, multiexponential T2 values were shown to depend on estimated axon diameter, myelin thickness, and inter-compartmental exchange (Dula et al., 2010). Differences in T2 mapping techniques and lack of gold standard prevent direct comparisons to our results, but support the observation that the variation in T2 intrinsically depends on the underlying tissue mesoscopic structure. In healthy controls, EMC T2 mapping demonstrated remarkably consistent differences amongst deep gray nuclei structures – for both automated and manual ROI analysis there was a 40% variation in T2 values between globus pallidus and caudate nucleus (Table 2).

White matter T2 values in healthy controls demonstrated spatial anatomic dependence (Fig. 5). In the cerebral hemispheres, T2 values for periventricular, centrum semiovale, subcortical and juxtacortical white matter decreased in a centrifugal manner (mean T2 decreasing 15% from 66.6 to 56.3 ms). A provocative result was that the T2 of juxtacortical white matter in the hand knob of the precentral gyrus appeared significantly higher than other juxtacortical regions (p < 0.0001). Further, the posterior limb of the internal capsule demonstrated the highest white matter T2 values (70.5 ± 3.9 ms) and was anecdotally noted to be brightest in the predicted location of the hand fibers for the corticospinal tract. Both these regions should include the corticospinal tract involved in the primary motor control of hand movements. Discordant T2 decreases observed in the posterior limb internal capsule for RRMS patients (Table 2) might reflect a different tissue environment and requires further investigation. Similarly, T2 for the genu, body and splenium of the corpus callosum, regions with different anatomic and functional connectivity, varied by 30% in healthy controls with all three regions statistically different (unpaired t-tests, p < 0.0001). Future studies could investigate whether T2 is affected by callosal or internal capsule anatomic regional differences in bound water fraction, axon size, density, dispersion, myelination, and orientation relative to the main magnetic field (B0). Accurate mapping of multiple T2 components may be particularly informative for these interesting observations.

Validation of EMC based T2 mapping

We have demonstrated the potential sensitivity of the novel EMC T2 mapping technique in a large cohort of RRMS patients. The EMC algorithm offers more accurate T2 mapping compared to conventional exponential fitting, which overestimated T2 values by 30–100%. Perhaps more important, EMC based T2 values are stable across different MRI scanners or acquisition protocols and does not have specific hardware requirements as recent advanced diffusion strategies (Ben-Eliezer et al., 2015b). Thus, it can be adapted to most existing MRI systems with multiecho acquisition capability. The variance amongst control subjects for individual regions was < 5%. Further, the coefficient of variation was < 4% for manual or automated ROI segmentation with repeat MRI scans in 3 healthy controls. Note that the EMC method differs substantially from previous T2 mapping approaches yet has been thoroughly validated against single spin echo reference acquisitions (Ben-Eliezer et al., 2015a, Ben-Eliezer et al., 2016). These methodological differences limit our ability to make direct comparisons with previous results in the MS patient population.

Conclusions

We report the first application of the EMC quantitative T2 mapping technique to RRMS patients. The current results demonstrate that this new method is stable and precise when applied to a large cohort of clinical subjects. We observed statistically significant T2 increases in both normal-appearing white and gray matter structures for subjects with a clinically-established diagnosis of RRMS. Similar to diffusion, magnetization transfer, and spectroscopy (Ceccarelli et al., 2007, Davie et al., 1997, Mangia et al., 2014), the observed T2 changes in normal-appearing tissues may reflect early neurodegeneration and/or chronic inflammatory myelin and glial pathology not visible to conventional MRI. In this initial case-control study, thalamic T2 values alone demonstrated excellent discrimination of RRMS patients from controls (AUC = 0.913). We also observed interesting T2 differences for different gray and white matter regions for healthy controls that may reflect functional and/or mesoscopic structural differences. Future prospective efforts will determine how region-specific T2 quantification can be used to improve the diagnostic accuracy for early MS or distinguish MS subtypes, and determine whether regional brain T2 quantification provide biomarkers for MS-associated clinical disability and disease progression.

Financial disclosure and declaration of interest

Timothy Shepherd received research support from the National Institute of Aging (NIH 1K23 AG048622-01). Ivan Kirov received research support from the National Institute of Neurological Disorders and Stroke (NIH R01 NS097494-01). This research was supported in part by the Center for Advanced Imaging Innovation and Research, an NIH NIBIB Biomedical Technology Resource Center (Grant P41EB017183). The authors have no other conflicts of interest to declare.

Table A1

Stability test results for the EMC vs. Exponential fitting. Columns 2–5: present inter- and intra-scanner variability of T2 values for 5 brain ROIs of a single healthy volunteer, collected over a series of 24 separate scans (see text). Columns 6–7: inter subject variability of the mean T2 values, estimated across a group of 38 healthy controls [EMC: EMC fitting; EXP: Exponential fitting].

ROI name	Intra-scanner variability [ms]		Inter-scanner variability [ms]		Inter-subject variability [ms]
	EMP	EXP	EMP	EXP	EMC	EXP
Genu of CC	1.5	1.5	0.6	1.1	2.3	3.9
Splenium of CC	1.4	1.7	0.6	1.1	3.0	5.7
Caudate nucleus	1.6	2.2	1.0	2.8	2.8	5.1
Juxtacortical WM	2.0	2.2	0.6	0.8	2.0	3.1
Periventricular WM	1.5	2.4	0.7	1.0	2.3	3.5
Average variability	1.6	2.0	0.7	1.4	2.5	4.3