Diffusion Basis Spectrum Imaging and Diffusion Tensor Imaging Predict Persistent Black Hole Formation in Multiple Sclerosis

Abstract

Background and Objectives:

Diffusion basis spectrum imaging (DBSI) extracts multiple anisotropic and isotropic diffusion tensors, providing greater histopathologic specificity than diffusion tensor imaging (DTI). Persistent black holes (PBH) represent areas of severe tissue damage in multiple sclerosis (MS), and a high PBH burden is associated with worse MS disability. This study evaluated the ability of DBSI and DTI to predict which acute contrast-enhancing lesions (CELs) would persist as T₁ hypointensities (i.e. PBHs) 12 months later. We expected that a higher radial diffusivity (RD), representing demyelination, and higher DBSI-derived isotropic non-restricted fraction, representing edema and increased extracellular space, of the acute CEL would increase the likelihood of future PBH development.

Methods:

In this prospective cohort study, relapsing MS patients with ≥1 CEL(s) underwent monthly MRI scans for 4 to 6 months until gadolinium resolution. DBSI and DTI metrics were quantified when the CEL was most conspicuous during the monthly scans. To determine whether the CEL became a PBH, a follow-up MRI was performed at least 12 months after the final monthly scan.

Results:

The cohort included 20 MS participants (median age 33 years; 13 women) with 164 CELs. Of these, 59 (36%) CELs evolved into PBHs. At Gd-max, DTI RD and AD of all CELs increased, and both metrics were significantly elevated for CELs which became PBHs, as compared to non-black holes (NBHs). DTI RD above 0.74 conferred an odds ratio (OR) of 7.76 (CI 3.77-15.98) for a CEL becoming a PBH (AUC 0.80, CI 0.73-0.87); DTI axial diffusivity (AD) above 1.22 conferred an OR of 7.32 (CI 3.38-15.86) for becoming a PBH (AUC 0.75, CI 0.66-0.83). DBSI RD and AD did not predict PBH development in a multivariable model. At Gd-max, DBSI restricted fraction decreased and DBSI non-restricted fraction increased in all CELs, and both metrics were significantly different for CELs which became PBHs, as compared to NBHs. A CEL with a DBSI non-restricted fraction above 0.45 had an OR of 4.77 (CI 2.35-9.66) for becoming a PBH (AUC 0.74, CI 0.66-0.81); a CEL with a DBSI restricted fraction below 0.07 had an OR of 9.58 (CI 4.59-20.02) for becoming a PBH (AUC 0.80, 0.72-0.87).

Conclusion:

Our findings suggest that greater degree of edema/extracellular space in a CEL is a predictor of tissue destruction, as evidenced by PBH evolution.

1.1.1. INTRODUCTION

In multiple sclerosis (MS), areas of acute inflammatory demyelination are associated with increased blood-brain barrier permeability creating gadolinium (Gd)-contrast enhancing lesions (CELs) on MRI.¹ After six or more months, 20-55% of acute CELs evolve into persistent black holes (PBHs).^{2, 3} Histopathologically, PBHs have greater axonal loss and expansion of the extracellular space compared to normal appearing white matter (NAWM) and to lesions that are non-black holes (NBHs).^{4, 5} Conversion of CELs into PBHs is associated with larger lesion size³ and longer duration of Gd-enhancement.² A high PBH burden is also associated with worse MS disability and disease progression,^6-8 and the evaluation, evolution, and quantification of PBHs is an important radiographic endpoint in MS studies.^9-11

Diffusion tensor imaging (DTI) assesses restriction and directionality of water movement to infer tissue microstructure.¹² In tightly packed axonal tracts in animal studies, DTI radial diffusivity (RD) increases in response to myelin and tissue injury, and DTI axial diffusivity (AD) decreases in response to axonal swelling and injury.¹³ In the presence of edema, crossing fiber tracts, or in areas with partial volume effects due to cerebrospinal fluid, DTI cannot fully discern demyelination and axon damage. Diffusion basis spectrum imaging (DBSI) overcomes some of these challenges,¹⁴ by modeling diffusion weighted MR signals as a linear combination of multiple anisotropic diffusion tensors, measuring RD and AD, and a spectrum of isotropic diffusion tensors, estimating restricted fraction (reflecting cellular density) and non-restricted fraction (reflecting extracellular space).¹⁵ Thus, DBSI has potential to provide greater histologic specificity than DTI.

We previously reported that CELs with a 40% elevation in DTI RD had a 5-fold increased risk of becoming a PBH.¹⁶ The goal of this study was to evaluate the ability of DBSI metrics to predict CEL evolution to PBHs and compare these findings to the predictive abilities of DTI. We speculated that edema confounded DTI AD and RD in CELs which could have reduced the specificity of DTI RD to demyelination in our prior study.¹⁴ Thus, in this study we hypothesized that DBSI metrics would also predict the development of PBH, but with greater specificity for the individual roles of vasogenic edema, demyelination, and axonal damage, as reflected in elevated DBSI isotropic non-restricted fraction, elevated DBSI RD, and decreased DBSI AD, respectively.

2.1.1. METHODS

2.2.1. Setting, consent, and participant profile

This study was approved by the Washington University Human Research Protection Office/Institutional Review Board and all participants provided written informed consent. The study was conducted at Washington University in St. Louis, Missouri, USA; recruitment occurred from June 2014 to March 2018 and data collection concluded June 2019. Inclusion criteria included adults with clinically definite relapsing MS by 2010 McDonald criteria¹⁷ and one or more CELs on routine, clinical brain MRI. Exclusion criteria included participants without a brain CEL on their first study MRI, pregnancy or breastfeeding, and contraindications to MRI. Treatment of MS continued per standard of care, irrespective of study participation.

2.3.1. Study protocol

Demographics and the Expanded Disability Status Scale (EDSS) were obtained at baseline to describe the cohort characteristics and disability levels, respectively.¹⁸ MS disease and medication history were obtained at baseline and each subsequent study visit. Participants underwent at least four monthly MRI scans at months 0, 1, 2 and 3 to determine time of maximum Gd-enhancement and time of Gd-resolution. If Gd-resolution was not observed by month 3, participants underwent additional scans at months 4, and 5 (as needed) until Gd-resolution was observed. An individual participant could have as many as six monthly scans. A follow-up MRI was performed at least 12 months after the final monthly scan, to classify CEL outcome as either PBH or NBH.

2.4.1. MRI protocol

2.4.2. Imaging acquisition

Participants were imaged on a 3.0 Tesla MRI scanner (Siemens Trio, Siemens Medical Solutions, Germany) using a 32-channel head coil. T₁W MPRAGE scans were done at isotropic 1×1×1 mm³ resolution for identification of structural landmarks and as a registration target (TR=2400 ms, TE=3.16 ms, TI=1000 ms, Matrix=256×224, FOV=256×224 mm², Resolution=1×1×1 mm³). Fluid attenuated inversion recovery (FLAIR) scans were acquired to quantify visible white matter lesion volumes (TR=7500 ms, TI=2500 ms, TE=210 ms, Matrix=256×256, FOV= 256×256 mm², Resolution=1×1×1 mm³). Diffusion-weighted images (DWIs) were collected with a 99-direction multi-b value diffusion encoding scheme by single-shot spin-echo (SE) echo-planar imaging sequence with the following key parameters: voxel size = 2×2×2 mm³; Maximum b-value = 1500 s/mm²; acquisition time = 15 minutes. Two transverse nondiffusion weighted image (b = 0 s/mm²) with opposite polarity of the phase encoding direction (AP and PA) were acquired for image processing. Gadoterate meglumine was given at 0.2 mL/kg (0.1 mmol/kg) with a 20-minute delay before acquiring a T1W MPRAGE post-contrast scan.

DWIs were pre-processed to minimize motion and susceptibility artifact due to non-zero off-resonance fields. The FSL TOPUP,¹⁹ which estimates the susceptibility-induced off-resonance field from pairs of images with reversed phase-encode blips and corrects susceptibility distortions, and FSL EDDY tool,²⁰ which corrects images with eddy current distortions and motion artifacts, were used. All FLAIR, DWI and follow-up MPRAGE scans were registered to MPRAGE of the first scan using an affine registration by FSL FLIRT.²¹ Both DTI and DBSI metric maps were estimated on the preprocessed DW images using the in-house software developed with MATLAB.¹⁵

2.4.3. Region of interest analysis

A region of interest (ROI) representing each CEL at the time of maximum Gd-enhancement (Gd-max) was visually determined and manually drawn on the T₁W MPRAGE image post-contrast using Amira v6.7 (Visage Imaging) by raters blinded to PBH outcome (LW, GA). If the CEL was persistent on multiple monthly scans, the ROI was drawn on the T₁W scan where the Gd-enhancement was most hyperintense, referred to as “Gd-max”. We determined ROIs at Gd-max instead of the first research scan at which a CEL was seen because a subset of ROIs increased in size and intensity over time, and we wanted to capture the CEL at the timepoint with the most inflammation. ROIs were drawn conservatively so as to exclude potential contamination with gray matter and ventricles. Areas of edema (i.e. T1 hypointensity) surrounding the contrast enhancement were not included in the ROI. However, the T₁W hypointensity within a ring-enhancing lesion was considered part of the lesion and included. CELs in the cerebellum and brainstem were excluded due to difficulty drawing ROIs confined to white matter only.

2.4.4. Persistent black hole determination

A PBH was defined as a hypointense region on T₁W MPRAGE image at the site of the initial CEL ROI that was present on a follow-up scan performed 12-15 months after the last CEL ceased enhancing. Two neurologists (LW, VL) independently assessed if ROIs developed into a PBH. A third adjudicator (AHC) decided any discrepancy between raters (n=15, 9% of total CELs). Raters were blinded to diffusion results and other neurologists’ ratings.

2.5.1. Statistical considerations

We estimated a required sample size of at least 25 participants with 50 CELs based on prior DTI data which showed that an increase of ^~5% in RD or AD, or a decrease of ^~5% in FA, increased PBH probability by ^~4%. For DTI, 50 CELs would have ≥ 0.80 power at .05 significance level to detect odds ratios (PBH/NBH) of 1.36, 2.15, and 0.278 for RD, AD, and FA, respectively. We estimated that the greater sensitivity and specificity of DBSI would provide greater power.

Demographics and lesion characteristics were summarized using descriptive statistics; mean (standard deviation [SD]) and median were used for continuous variables, as appropriate, and frequency (percentage) used for categorical variables. Clustered logistic regression was used to determine association of DBSI and DTI metrics on the risk of a lesion progressing to a PBH, with covariates of lesion volume³ and months of Gd-enhancement² used in multivariable models, based on prior research.

Additionally, a receiver operating characteristic curve analysis was conducted to identify a threshold for each DBSI metric at Gd-max that predicted PBH status using logistic regression. The threshold was determined using the Youden’s index and the area under the curve (AUC) and its 95% confidence interval for the model reported. Following the identification of the threshold, the association of that DBSI metric threshold at Gd-max was evaluated using binary logistic regression and odds ratio and their 95% confidence intervals of the association are reported. Statistical analyses were conducted in SAS v9.4 and a significance level of 0.05 was used.

3.1.1. RESULTS

3.2.1. Participant demographics and lesion characteristics:

Twenty-six participants enrolled, and 20 participants (77%) completed study procedures and were included in analysis. Two subjects were excluded due to excessive motion artifact, two were lost to follow up, one did not have any CELs on the first MRI, and one was excluded due to uncertain diagnosis. Participants included in the analysis had a median age of 33 years (range 20 – 55 years), a median MS disease duration of 2.12 years (range 0.01 – 24.75), and a median baseline EDSS of 2 (range 1 – 5.5) (Table 1). Sixteen (80%) participants reported a change in their disease modifying therapy (DMT) during the study period (Supplemental Table A).

Table 1.

Baseline participant demographics.

Characteristics	Values
Number of participants	20
Age, years, median (range)	33 (20 – 55)
Sex, female, n (percentage of total participants, %)	13 (65%)
Race, n
White, not Hispanic	13
Hispanic	1
Black	6
MS Disease Duration, years, median (range)	2.12 (0.01 – 24.75)
Baseline EDSS, median (range)	2 (1 – 5.5)
Baseline DMT, n
glatiramer acetate	3
interferon	1
teriflunomide	1
dimethyl fumarate	3
fingolimod	1
natalizumab	2
alemtuzumab	1
none	8
Number of participants who received steroids during period of monthly scans, n (percentage of total participants, %)	10 (50%)

Abbreviations: MS = multiple sclerosis; EDSS = Expanded Disability Status Scale; DMT = disease modifying therapy

In this study, 195 CELs were identified and 164 CELs met criteria for inclusion. Reasons for exclusion of the 31 CELs were: CEL present in gray matter (n=14), questionable / faint enhancement (n=16), and lesion not demonstrated on FLAIR imaging (n=1).

Of the 164 CELs that met criteria for analysis, 59 (36%) evolved into a PBH. Participants had a median of 8 CELs with median enhancement duration of 1 month (i.e. present on a single monthly scan) (Table 2). Median time between the last monthly scan and follow-up scan to determine PBH outcome was 12.4 months (range 11.7-19.2). Fifty-five CELs (33.5% of those analyzed) were at Gd-max on the first scan on which they were seen, 38 CELs (23.2%) were at Gd-max one month after the scan on which they were first seen, 43 CELs (26.2%) were at Gd-max two months after the scan on which they were first seen, and 28 CELs (17.1%) were at Gd-max at least three months after the scan on which they were first seen.

Table 2.

Contrast-Enhancing Lesion Characteristics.

Contrast-Enhancing Lesion Characteristics (n=164)	Values
Number of CELs per participant, median (25th, 75th percentile)	8 (4, 15)
Duration of Gd-enhancement, months, median (SD)	1 (0.74)
Number of PBHs, n (percentage of total CELs, %)	59 (36%)
Number of monthly scans per participant, median (range)	4.5 (4-6)
Monthly Scan at which a CEL was first observed, n (percentage of total CELs, %)
0	63 (38.4%)
1	44 (26.8%)
2	32 (19.5%)
3	10 (6.1%)
4	7 (4.3%)
5	8 (4.9%)
Time between monthly and final follow-up scan, months, median (range)	12.4 (11.7-19.2)

Abbreviations: CELs = contrast-enhancing lesions; Gd = gadolinium; PBH = persistent black hole

3.3.1. Diffusion tensor imaging radial and axial diffusivity predicted persistent black hole development

DBSI and DTI metrics of the CEL ROI were assessed up to 3 months before and 5 months after Gd-max (Figure 1). DTI AD and RD of PBHs and NBHs increased at Gd-max; DTI RD (but not AD) for PBHs was significantly elevated at each time point, as compared to NBHs (Figure 1A).

Figure 1. Diffusion imaging over time in relation to maximum gadolinium enhancement (Gd-max).

Gd-max = month 0. A: Diffusion tensor imaging (DTI) axial diffusivity (AD) and radial diffusivity (RD) of lesions that became persistent black holes (PBHs) and non-black holes (NBHs). B: Diffusion basis spectrum imaging (DBSI) AD and RD of lesions that became PBHs and NBHs. C: DBSI non-restricted fraction (non-RF) and restricted fraction (RF) of lesions that became PBHs and NBHs. The number of CELs which became PBHs and NBHs is noted below each timepoint. As some lesions developed during the study and reached Gd-max after initial observation, the graph shows diffusivity assessed up to 3 months before and 5 months after Gd-max (i.e. their peak intensity). * indicates statistically significant differences between PBHs and NBHs at Gd-max.

PBH	9	15	35	59	57	51	40	30	16
NBH	16	41	66	105	98	91	78	43	21

(See “Figure 1” document for figure)

In a univariate logistic model, increases of DTI AD and RD and decreases of DTI FA distinguished lesions that would become a PBH (Table 3). However, in a multivariable model controlling for lesion size and duration of Gd-enhancement, the ability of DTI FA to predict PBH formation was no longer statistically significant.

Table 3.

Clustered Logistic Regression for predicting persistent black hole status based on diffusion tensor imaging (DTI) and diffusion basis spectrum imaging (DBSI) metrics.

	Univariate Model			Multivariable Model
Metric	Odds Ratio	95% CI	p-value	Odds Ratio	95% CI	p-value
DTI Axial Diffusivity	1.70	1.36-2.11	<.0001	1.36	1.08-1.71	0.009
DTI Radial Diffusivity	1.82	1.48-2.24	<.0001	1.47	1.16-1.87	0.002
DTI Fractional Anisotropy	0.45	0.30-0.65	<.0001	0.69	0.47-1.01	0.06
DBSI Axial Diffusivity	1.13	0.99-1.29	0.06	1.03	0.89-1.19	0.71
DBSI Radial Diffusivity	1.75	1.34-2.29	<.0001	1.22	0.90-1.63	0.19
DBSI Fractional Anisotropy	0.46	0.30-0.70	0.0005	0.70	0.45-1.10	0.12
DBSI Restricted Ratio	0.12	0.05-0.31	<.0001	0.23	0.10-0.56	0.001
DBSI Non-restricted Ratio	1.75	1.39-2.21	<.0001	1.34	1.06-1.73	0.02

Abbreviations: CI = confidence interval; Univariate and Multivariable model results are shown. The multivariable model controls for the number of voxels in the CEL (size) and duration of enhancement. Statistically significant p-values are bolded and italicized.

Within the present cohort, a CEL with a DTI RD above 0.74 had an OR of 7.76 (95% CI 3.77-15.98) for becoming a PBH (AUC 0.80; 95% CI 0.73-0.87), consistent with our prior published study.¹⁶ A CEL with a DTI AD above 1.22 had an OR of 7.32 (95% CI 3.38-15.86) for becoming a PBH (AUC 0.75; 95% CI 0.66-0.83).

3.4.1. Diffusion basis spectrum imaging isotropic metrics (restricted and non-restricted fraction) predicted persistent black hole development

Over time, DBSI RD (but not AD) of PBHs was significantly elevated at most timepoints, as compared to NBHs (Figure 1B). Restricted and non-restricted fractions are derived from the same DWI data and have an inverse relationship. Therefore, at Gd-max, restricted fraction decreased and DBSI non-restricted fraction increased in PBHs and NBHs. Non-restricted fraction of PBHs and NBHs was significantly different at all timepoints (Figure 1C).

In a univariate logistic model, decreased DBSI FA, increased DBSI RD, decreased restricted isotropic fraction and increased non-restricted fraction each distinguished lesions that would become a PBH. AD determined by DBSI did not distinguish lesions that would become a PBH, consistent with our prior study using AD determined by DTI.¹⁶ In a multivariable model controlling for lesion size and duration of Gd-enhancement, statistical significance of the predictive ability of anisotropic DBSI metrics (i.e. RD and FA) for PBH formation was lost (Table 3).

Isotropic DBSI metrics (i.e. restricted and non-restricted fraction) continued to predict future PBH formation in the multivariable model. A CEL with a DBSI non-restricted fraction above 0.45 had an OR of 4.77 (95% CI 2.35-9.66) for becoming a PBH (AUC 0.74; 95% CI 0.66-0.81) and a CEL with a DBSI restricted fraction below 0.07 had an OR of 9.58 (95% CI 4.59-20.02) for becoming a PBH (AUC 0.80; 95% CI 0.72-0.87).

4.1.1. DISCUSSION

In this study, we used PBHs as a surrogate for histopathology of chronic axonal loss and tissue damage.^{4, 5} We assessed the severity of tissue injury during the evolution of CELs using DTI and DBSI, and examined whether DTI and DBSI metrics might predict which CELs would become PBHs. We found that CELs with higher DTI AD and RD were more likely to develop into PBHs. However, DTI AD and RD represent the average of both isotropic and anisotropic tensors within tissues, meaning that DTI metrics are confounded by cellular inflammation, edema, increased extracellular space, gliosis, crossing fibers, and partial volume contamination.²² In contrast, DBSI detects inflammation through modeling to differentiate isotropic from anisotropic diffusion signals; DBSI is, therefore, more histopathologically specific.^{15, 23, 24} In our study, isotropic metrics (DBSI restricted and non-restricted fraction) were superior to DBSI anisotropic metrics in predicting PBH evolution, even after controlling for duration of Gd-enhancement and lesion size. Our findings indicate that excess extra-fiber isotropic diffusion signal in the newly forming CEL may directly predict the tissue destruction and is a key biomarker of PBH evolution. The results also supported our suspicion that the predictive abilities of increased DTI RD and AD for PBH formation are in large part due to the elevated isotropic component (i.e., edema and/or tissue loss) contributing to these DTI-derived metrics.

Because non-restricted fraction can approximate the level of edema in a CEL, DBSI has unique advantages over DTI metrics that reflect a summation of both anisotropic and isotropic diffusion. DBSI isotropic metrics may also be of clinical interest for the noninvasive quantitation of edema and areas of increased extracellular space and of cellularity within the CNS in other neurological and neurosurgical diseases. Conversely, DTI metrics are simpler, making calculations easier and DTI more practical in a clinical setting.

DBSI restricted fraction putatively reflects cellularity. This is supported by the strong correlations between DBSI restricted fraction with cell nuclei determined by immunohistochemistry in mice and in human autopsy and biopsy specimens.²³ DBSI restricted fraction does not, however, distinguish between intrinsic and extrinsic cellularity. In the present study, DBSI restricted fraction decreased during CEL formation and a DBSI restricted fraction below 0.07 conferred an almost 10-fold higher risk of becoming a PBH. We suspect that increased extracellular space in new lesions (primarily due to acute vasogenic edema and/or axonal loss) resulted in the dilution of cellular density driving the DBSI restricted fraction lower. Other modalities, such as positron emission tomography (PET) with specific radioligands, are more adept than DBSI at identifying and quantifying specific molecular and cellular components. For example, translocator protein (TPSO) PET can detect microgliosis and infiltrating macrophages,²⁵ but this technique is limited by diffuse expression of TPSO and genetic polymorphisms that affect binding affinity. Newer PET tracers are being developed to identify other specific components of neuroinflammation such as lymphocytes,²⁶ inducible nitric oxide synthase²⁷ or CD20 B cells²⁸ but each of these ligands binds only one histopathologic component. Advances in PET imaging overall have improved its temporal²⁹ and spatial resolution,³⁰ but PET imaging of the CNS still lags far behind the resolution of MRI. PET also uses ionizing radiation. Therefore, DBSI may be more feasible than PET for larger-scale studies and longitudinal studies involving frequent scans.

Previously, our group reported that a CEL with a DTI RD ratio (lesion RD to contralateral NAWM RD) of ≥1.4 was 5 times more likely to become a PBH.¹⁶ The present study reports consistent relationships between PBH development and increased DTI-derived RD, but we used absolute lesion RD instead of a ratio because of the abundance of MS lesions in the contralateral hemispheres, and the known frequent histopathologic abnormalities below the resolution of MRI in the NAWM.³¹ Notably, another group found the same relationship between DTI RD and PBH formation in 21 participants with 60 CELs.³² In the current study, DBSI RD was not a predictor of PBH formation after controlling for duration of Gd-enhancement and lesion size, unlike DTI RD. This is likely because DTI RD reflects not only anisotropic diffusion, but is also confounded by isotropic diffusion components. In the setting of an acute CEL, DTI RD will increase due to the isotropic diffusion increase from edema combined with an increase in RD due to demyelination. Increased DBSI RD excludes isotropic diffusion and is more reflective of demyelination. DBSI RD was not predictive in the multivariable model, suggesting that acute demyelination has minimal impact on subsequent PBH formation.

Animal studies suggest that DTI AD decreases in the setting of axonal injury³³ yet changes in DTI-derived AD have not been predictive of PBH formation in people with MS.^{16, 32, 34} In contrast to our prior results, here we found that increased DTI AD predicted PBH formation.¹⁶ This discrepancy may be because axon injury will decrease DTI AD, but axon loss increases the extracellular space within a white matter tract, increasing DTI AD.²³ Loss of fiber tract organization has been noted in response to neural injury in other human DTI studies and can confound DTI AD.^{35, 36} Because DTI-derived metrics are a composite of both anisotropic and isotropic diffusion measures, DTI-derived AD may be increased completely independent of axonal pathology.¹⁵

One strength of this study was the inclusion of DBSI and DTI, which allowed us to explore the advantages and disadvantages of each modality within a single, co-registered dataset. Our statistical analyses also controlled for duration of Gd-enhancement² and lesion size,³ each of which could influence PBH evolution. Finally, compared to similar studies exploring predictors of PBH formation, this study, with 164 CELS, included an equivalent or larger number of lesions.^{3, 16, 37}

One limitation of our study was that there were only 20 participants in the study and 50% of the CELs came from five participants. This is potentially problematic because actively demyelinating lesion histopathology is thought to be more heterogeneous between patients than within the same patient which may limit generalizability.³⁸ Second, the borders of active lesions are often ill-defined.³⁸ Therefore, partial volume averaging, though reduced by DBSI in comparison to DTI, may have affected voxels at the borders of our lesions. Also, 33.5% of CELs were deemed to be at Gd-max at the first timepoint, so we could not definitively rule out that these CELs might have already been past their maximum intensity. Potentially, this could result in less inflammation and changes in the DBSI and DTI values for these lesions. However, considering that the median duration ROI enhancement was one month, consistent with prior literature,³⁹ we expect that this did not significantly impact our results.

Due to limited sample size, varied onset and number of CELs in each participant, and frequent DMT changes, we were not able to assess the relationship between DBSI metrics and clinical outcomes or DMT. However, other groups have found that DBSI metrics distinguish differences between NAWM of various MS subtypes, correlate with cognitive dysfunction and motor impairment, and are associated with serum neurofilament light chain levels.^{31, 40-43} These relationships should continue to be explored in future studies. Finally, we were also unable to examine the relationship between steroids and DBSI metrics because of the variability in steroid administration. The evidence for the impact of steroids on PBH evolution is currently mixed. Steroid administration after optic neuritis does not change one⁴⁴ or five-year⁴⁵ visual function or degree of axonal damage in optic nerves after induction of murine experimental autoimmune encephalitis (EAE). However, a small randomized study suggested monthly high-dose steroids over a period of five years may slow the development of PBHs⁴⁶.

4.2.1. Conclusions

In summary, we explored if DBSI and DTI metrics within new MS lesions could predict which CELs would evolve to become PBHs, signifying worse tissue destruction. We found that changes in isotropic DBSI metrics (increased non-restricted isotropic fraction, reduced restricted isotropic fraction) predicted PBH formation and we confirmed prior reports of increased DTI RD predicting PBH formation. However, in view of the DBSI results, we now interpret this increase in RD derived from DTI as reflecting primarily increased edema, and not solely demyelination. In the new era of precision medicine, DBSI could provide a clinically and histopathologically relevant way to precisely track individual patient responses to therapy or serve as a pathology surrogate in future remyelination and neuroprotection trials.

Supplementary Material

1

Supplemental Table A. Number of Contrast Enhancing Lesions, Persistent Black Holes, Steroid Administration, and Disease Modifying Therapies by Month for each Participant. CEL = contrast enhancing lesion; PBH = persistent black hole; DMT = disease modifying therapy; GA = glatiramer acetate; DMF = dimethyl fumarate; IFN = interferon; TFM = teriflunomide; ALM = alemtuzumab; NAT = natalizumab; S1P = S1P modulator; *OTHER = treated with alemtuzumab previously. Participants received high-dose steroids in months shaded in gray.

HIGHLIGHTS:

• Diffusion MRI measures the effect of tissue microstructures on Brownian motion of water to assess histopathology

• DTI radial diffusivity (reflecting demyelination and edema) predicts future tissue damage in MS

• Diffusion basis spectrum imaging (DBSI) models multiple diffusion tensors, improving pathologic specificity

• DBSI isotropic tensors (representing cellularity and edema) predict future tissue damage in MS

• DBSI may provide additional histopathologic insights in acute MS lesions