Feasibility of postoperative diffusion-weighted imaging to assess representations of spinal cord microstructure in cervical spondylotic myelopathy

Abstract

OBJECTIVE

Diffusion basis spectrum imaging (DBSI) has shown promise in evaluating cervical spinal cord structural changes in patients with cervical spondylotic myelopathy (CSM). DBSI may also be valuable in the postoperative setting by serially tracking spinal cord microstructural changes following decompressive cervical spine surgery. Currently, there is a paucity of studies investigating this topic, likely because of challenges in resolving signal distortions from spinal instrumentation. Therefore, the objective of this study was to assess the feasibility of DBSI metrics extracted from the C3 spinal level to evaluate CSM patients postoperatively.

METHODS

Fifty CSM patients and 20 healthy controls were enrolled in a single-center prospective study between 2018 and 2020. All patients and healthy controls underwent preoperative and postoperative diffusion-weighted MRI (dMRI) at a 2-year follow-up. All CSM patients underwent decompressive cervical surgery. The modified Japanese Orthopaedic Association (mJOA) score was used to categorize CSM patients as having mild, moderate, or severe myelopathy. DBSI metrics were extracted from the C3 spinal cord level to minimize image artifact and reduce partial volume effects. DBSI anisotropic tensors evaluated white matter tracts through fractional anisotropy, axial diffusivity, radial diffusivity, and fiber fraction. DBSI isotropic tensors assessed extra-axonal pathology through restricted and nonrestricted fractions.

RESULTS

Of the 50 CSM patients, both baseline and postoperative dMR images with sufficient quality for analysis were obtained in 27 patients. These included 15 patients with mild CSM (mJOA scores 15–17), 7 with moderate CSM (scores 12–14), and 5 with severe CSM (scores 0–11), who were followed up for a mean of 23.5 (SD 4.1, range 11–31) months. All preoperative C3-level DBSI measures were significantly different between CSM patients and healthy controls (p < 0.05), except DBSI fractional anisotropy (p = 0.31). At the 2-year follow-up, the same significance pattern was found between CSM patients and healthy controls, except DBSI radial diffusivity was no longer statistically significant (p = 0.75). When assessing change (i.e., postoperative − preoperative values) in C3-level DBSI measures, CSM patients exhibited significant decreases in DBSI radial diffusivity (p = 0.02), suggesting improvement in myelin integrity (i.e., remyelination) at the 2-year follow-up. Among healthy controls, there was no significant difference in DBSI metrics over time.

CONCLUSIONS

DBSI metrics derived from dMRI at the C3 spinal level can be used to provide meaningful insights into representations of the spinal cord microstructure of CSM patients at baseline and 2-year follow-up. DBSI may have the potential to characterize white matter tract recovery and inform outcomes following decompressive cervical surgery for CSM.

CERVICAL spondylotic myelopathy (CSM) is a chronic, progressive degenerative disease of the cervical spine.¹ Although surgical decompression remains the mainstay treatment for moderate and severe myelopathy, long-term outcomes are unpredictable.² In recent years, advanced diffusion-weighted MRI (dMRI) techniques capable of extracting quantitative metrics of spinal cord microstructure, such as diffusion basis spectrum imaging (DBSI),^3,4 have emerged to better evaluate CSM patients.⁵ We have previously shown that DBSI provides meaningful prognostic value when predicting long-term clinical outcomes in CSM.^6,7 However, the ability of DBSI to track spinal cord integrity after decompressive surgery remains unknown.

Extracting dMRI data from postoperative spine images is particularly challenging. Spinal instrumentation introduces significant image artifact, and as such, few studies have assessed longitudinal dMRI techniques,^8–11 and none have assessed postoperative DBSI. Given this evidence gap, we investigated the feasibility of using DBSI postoperatively, focusing on the rostral C3 spinal cord level to minimize metal artifact and maximize imaging homogeneity. In addition, we examined the ability of DBSI metrics derived from the C3 level alone to assess changes in spinal cord microstructure on baseline evaluations. Furthermore, we include a healthy control population to assess the long-term repeatability of DBSI and evaluate whether C3-level DBSI metrics can differentiate CSM patients from healthy controls.

Methods

Fifty CSM patients and 20 healthy controls were prospectively enrolled by a single tertiary care institution between 2018 and 2020. Inclusion and exclusion criteria are described in Supplemental Table 1. At initial presentation, demographic information, symptom duration, and a complete neurological examination were performed. Patients were characterized as having mild (modified Japanese Orthopaedic Association [mJOA] scores 15–17), moderate (mJOA scores 12–14), or severe (mJOA scores 0–11) myelopathy based on established mJOA classifications.¹² Consistent with previous studies, the minimal clinically important difference (MCID) of the mJOA score was defined as a 2-point improvement.¹³ All CSM patients underwent decompressive cervical surgery, with either an anterior (i.e., anterior cervical discectomy and fusion [ACDF] or arthroplasty) or posterior (posterior cervical discectomy and fusion or laminoplasty) approach. The surgical approach was selected based on surgeon discretion. Both CSM patients and healthy controls were followed longitudinally for up to 2 years postoperatively. This study was designed and reported according to STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines¹⁴ and approved by our institution’s IRB. Informed consent was obtained for all patients.

Diffusion Basis Spectrum Imaging

MRI and dMRI data were acquired using a 3T Prisma scanner (Siemens) (see Supplemental Methods for MRI data acquisition parameters). DBSI models intravoxel non-Gaussian diffusion processes as a linear combination of anisotropic and isotropic Gaussian diffusion tensors. DBSI resolves co-occurring pathologies such as demyelination, typically associated with changes in anisotropic diffusion, and the incident immune response, reflected by isotropic diffusion (Supplemental Table 2). The DBSI metrics corresponding to anisotropic diffusion are 1) fiber fraction, a measure of apparent fiber density; 2) fiber fractional anisotropy, which reflects tract integrity; and 3) axial diffusivity and 4) radial diffusivity, which quantify diffusion parallel and perpendicular to a fiber bundle reflecting axonal and myelin integrity, respectively. The DBSI-measured extra-fiber isotropic apparent diffusion coefficient (ADC) is related to compartment size. Intracellular water diffusion leads to a small ADC reflecting the restricted diffusion fraction inside cells (ADC ≤ 0.3 μm²/msec). Meanwhile, nonrestricted isotropic ADC values reflect edematous water or CSF (ADC > 0.3 μm²/msec), which theoretically accounts for the nonrestricted diffusion fraction from noncellular sources in biological tissue (e.g., vasogenic edema and CSF).

Derivation of DBSI Metrics

Diffusion-weighted MRI signals of the spinal cord were processed using a Python image processing pipeline that was developed in-house. The pipeline is initiated by converting the imaging data from native DICOM to NIFTI (Neuroimaging Informatics Technology Initiative) format, and then proceeds to correct for motion artifacts derived from either physiological processes such as a cardiac activity or motion from the patient. After DBSI analysis of the diffusion-weighted images, hand-drawn regions of interest were applied to each participant’s DBSI-derived metric maps at the C3 to C6 spinal levels to extract estimated values from the right and left lateral motor, ventral motor, and dorsal sensory regions. In this study, DBSI metrics derived from the C3 level (at the midvertebral level) were included for analysis. Changes in DBSI metrics were calculated by subtracting preoperative values from postoperative values.

Statistical Analysis

Univariate statistics were used to analyze descriptive data. Patients were classified as healthy control or CSM subgroups based on mJOA classifications. Imaging data were found to be nonnormally distributed using the Shapiro-Wilk test. Preoperative, postoperative, and change in C3-level DBSI data were compared between healthy controls and CSM patients using independent-samples Wilcoxon rank-sum tests. To assess whether preoperative DBSI metrics were significantly different from postoperative measures within each group (i.e., control and CSM groups), pairwise Wilcoxon rank-sum tests were run. To examine the ability of C3-level data alone to characterize differences in spinal cord microstructure between healthy controls and CSM patients in the preoperative setting, spinal cord diffusion data from all available levels (i.e., C3 to C6) on baseline evaluation were combined and compared between groups. A similar analysis was not performed at the 2-year follow-up because of metallic artifact in postoperative CSM patients. In a sensitivity analysis to assess the effect of surgical level on imaging data, DBSI metrics were compared between patients undergoing surgeries including C3 versus those undergoing surgeries below C3.

Preoperative, postoperative, and change in DBSI metrics were compared between CSM patients who did and did not improve based on mJOA score. In addition, preoperative, postoperative, and change in DBSI metrics were correlated with the preoperative, postoperative, and change in mJOA scores, respectively, using Spearman correlations due to nonnormally distributed data. Last, preoperative, postoperative, and change in DBSI metrics were assessed among patients with different CSM severity based on mJOA classifications using ANOVA. A false discovery rate correction was applied for multiple comparisons testing, and statistical significance was set a priori at p < 0.05. All statistical analyses were performed in R version 4.1.4 (The R Foundation for Statistical Computing).

Results

Fifty CSM patients and 20 healthy controls were enrolled in the study (Fig. 1). All patients and healthy controls underwent preoperative and postoperative dMRI, followed by DBSI processing. In the preoperative period, 8 CSM patients and 1 healthy control were excluded due to MRI data of insufficient quality. At 2 years postoperatively, 9 CSM patients were lost to follow-up. After postoperative imaging, 6 CSM patients and 3 healthy controls were excluded due to poor MRI quality. The CSM patients excluded at the final follow-up had sufficient spinal instrumentation artifact that made data interpretation unreliable. Ultimately, 27 CSM patients and 16 healthy controls had complete clinical and dMRI data for analysis. This included 15 (56%) patients with mild CSM, 7 (26%) with moderate CSM, and 5 (19%) with severe CSM. The mean mJOA score for CSM patients was 14.3 (SD 3) compared with 17.5 (SD 1) in healthy controls (p < 0.001). The mean follow-up was 23.5 (SD 4.1, range 11–31) months.

FIG. 1.

Study flowchart. CON = controls.

Complete demographic, clinical, and surgical characteristics are described in Table 1. There were no significant differences in baseline demographics between CSM patients and healthy controls (all p > 0.05). Most CSM patients were male (67%), with a mean age of 56.7 (SD 7.7) years and a median symptom duration of 12 (range 2–240) months. Most surgeries were multilevel (89%), involving the C4–5 (16/53, 30%) or C5–6 (22/53, 42%) levels, with 17 (63%) patients receiving anterior surgery and 10 (37%) patients receiving posterior surgery. CSM patients experienced significant improvements in their mJOA score (mean change +1.3 [SD 1.9]) at the latest follow-up (p < 0.001). Based on an MCID of 2 points for change in the mJOA score, 10 patients improved, 16 were unchanged, and 1 worsened at the 2-year follow-up.

TABLE 1.

Demographic, clinical, and surgical characteristics

Characteristic	Healthy Controls (n = 16)	CSM Patients (n = 27)	p Value
Mean age, yrs	57.5 (8.5)	56.7 (7.7)	0.78
Mean BMI, kg/m2	27.1 (6.4)	29 (6.2)	0.36
Male sex	9 (56)	18 (67)	0.89
Caucasian race	12 (75)	21 (78)	0.74
Median (range) symptom duration, mos		12 (2–240)
Op type
Anterior		17 (63)
Posterior		10 (37)
Multilevel op		24 (89)
Instrumented levels treated		53
C1–2		0
C2–3		0
C3–4		6 (11)
C4–5		16 (30)
C5–6		22 (42)
C6–7		9 (17)
C7-T1		0
mJOA score
Preop	17.5 (1)	14.3 (3)	<0.001
Postop		15.6 (2.6)
Change		1.3 (1.9)*
Mean follow-up, mos		23.5 (4.1)

Values are given as number of patients or levels (%) or mean (SD) unless otherwise indicated. Boldface type indicates statistical significance.

^*Statistically significant on pairwise t-test (p < 0.001).

DBSI data were able to be effectively extracted from the C3 spinal cord level. Preoperative, postoperative, and change in DBSI measures are described in Tables 2, 3, and 4, respectively. DBSI metrics at all time points were not significantly different between patients undergoing surgeries including C3 (n = 6) compared with those undergoing surgeries below C3 (n = 21) (Supplemental Table 3). When comparing preoperative C3-level DBSI metrics between healthy controls and CSM patients, all measures were significantly different between groups (all p < 0.05) except fractional anisotropy (p = 0.31) (Table 2). In addition, a sensitivity analysis was performed to compare the ability of preoperative C3-level data alone versus that of combined averaged C3 to C6 data to differentiate healthy controls from CSM patients at baseline evaluation. When using combined C3- to C6-level data, a similar significance pattern was found, except with respect to DBSI radial diffusivity, which was no longer statistically significant (p = 0.06) (Supplemental Table 4).

TABLE 2.

Preoperative C3 slice DBSI metrics

	Healthy Controls (n = 16)	CSM Patients (n = 27)	p Value*
Fractional anisotropy	0.8517 (0.035)	0.8242 (0.067)	0.310
Fiber fraction	0.8596 (0.066)	0.7983 (0.072)	0.008
Axial diffusivity, μm2/msec	2.2475 (0.129)	2.6072 (0.312)	<0.001
Radial diffusivity, μm2/msec	0.2921 (0.054)	0.3853 (0.135)	0.025
Restricted fraction	0.0011 (0.003)	0.0295 (0.033)	0.006
Nonrestricted fraction	0.0121 (0.021)	0.0612 (0.084)	0.024

Values are presented as mean (SD). Boldface type indicates statistical significance.

^*Independent-samples Wilcoxon rank-sum tests.

TABLE 3.

Postoperative C3 slice DBSI metrics

	Healthy Controls (n = 16)	CSM Patients (n = 27)	p Value*
Fractional anisotropy	0.8244 (0.046)	0.8372 (0.057)	0.56
Fiber fraction	0.8769 (0.13)	0.7504 (0.085)	<0.001
Axial diffusivity, μm2/msec	2.2333 (0.16)	2.5426 (0.287)	<0.001
Radial diffusivity, μm2/msec	0.322 (0.066)	0.3286 (0.106)	0.75
Restricted fraction	0.0101 (0.021)	0.0507 (0.039)	<0.001
Nonrestricted fraction	0.0357 (0.107)	0.0596 (0.073)	0.006

Values are presented as mean (SD). Boldface type indicates statistical significance.

^*Independent-samples Wilcoxon rank-sum tests

TABLE 4.

Change in C3 slice DBSI metrics

	Healthy Controls (n = 16)	CSM Patients (n = 27)	p Value*
Fractional anisotropy	−0.0272 (0.052)	+0.013 (0.069)	0.065
Fiber fraction	+0.0173 (0.125)	−0.0479 (0.113)	0.053
Axial diffusivity, μm2/msec	−0.0142 (0.227)	−0.0645 (0.465)	0.551
Radial diffusivity, μm2/msec	+0.0299 (0.064)	−0.0566 (0.149)†	0.028
Restricted fraction	+0.0091 (0.022)	+0.0212 (0.053)	0.403
Nonrestricted fraction	+0.0236 (0.109)	−0.0016 (0.115)	0.783

Values are presented as mean (SD). Boldface type indicates statistical significance.

^*Independent-samples Wilcoxon rank-sum test.

^†Pairwise Wilcoxon rank-sum test.

On postoperative C3-level DBSI metrics, all measures were significantly different between groups (all p < 0.05) except fractional anisotropy (p = 0.56) and radial diffusivity (p = 0.75) (Table 3). When assessing change (i.e., postoperative − preoperative values) in C3-level DBSI measures, only radial diffusivity was significantly different between healthy controls and CSM patients (p = 0.028) (Table 4). Among CSM patients, there was a corresponding significant decrease in radial diffusivity (p = 0.02, pairwise Wilcoxon rank-sum test). Among healthy controls, there was no significant difference in DBSI metrics between pre- and postoperative time points (all p > 0.05, pairwise Wilcoxon rank-sum test).

When comparing C3-level DBSI metrics between CSM patients who improved and those who did not improve based on the MCID of the mJOA score, only the preoperative nonrestricted fraction and postoperative restricted fraction were significantly greater in CSM patients who did not improve (Supplemental Table 5). When correlating C3-level DBSI metrics with the mJOA score, only preoperative DBSI axial diffusivity (r_s = −0.34, p = 0.03) and preoperative DBSI restricted fraction (r = −0.49, p < 0.001) were significantly associated with preoperative mJOA score (Supplemental Table 6). There were no significant associations in the postoperative time points. Corresponding to these findings, there were no significant differences in C3-level DBSI metrics among mild, moderate, and severe CSM subgroups (Supplemental Table 7).

Discussion

Advancements in neuroimaging techniques have greatly aided clinical decision-making for patients with CSM.^12,15,16 In the postoperative setting, longitudinal dMRI may serve as a useful adjunct for the ability of dMRI to objectively assess representations of spinal cord microstructure. However, there is a paucity of studies assessing postoperative dMRI data, likely because of significant difficulties in attaining quantitative data of acceptable quality. Consequently, we investigated the clinical feasibility of longitudinal DBSI measurements, evaluating their stability over time and efficacy in differentiating the spinal cord integrity of CSM patients from that of healthy controls postoperatively.

Our results demonstrated that the C3 level was not only clinically feasible, but also sufficient to detect statistically significant differences in spinal cord integrity between CSM patients and healthy controls. In addition, the majority of DBSI metrics remained stable at the 2-year follow-up, suggesting that decompressive surgery was at least able to prevent further deterioration of spinal cord microstructure.

Rationale Behind C3 Segment Selection

In our imaging protocol, data from diffusion-weighted images are extracted from 7.5-mm slices at the midvertebral level from C3 to C6. Given that most of the surgeries in our cohort involved the C4–5 (30%) and C5–6 (42%) levels, data derived from the C4, C5, or C6 spinal levels exhibited significant image artifact, preventing adequate extraction of DBSI metrics (Fig. 2). In addition, the C1, C2, and C7 levels were not imaged. In our preliminary work, imaging these additional segments not only added a significant time burden for postoperative spine patients during scanning but also yielded suboptimal data. Specifically, these cervical levels are susceptible to partial volume effects, which occur when portions of different tissue types (e.g., white matter tracts and CSF) are averaged together in the image voxels of axial dMRI slices.^17–20 The C1 and C2 levels are subject to partial volume effects on standard brain dMRI, likely because of the natural coning of the spinal cord as it transitions from the brainstem to the cervical spine. The C7 level is similarly susceptible because of the natural curving of the spinal cord as it enters the thoracic spine. In addition, lower cervical regions such as C7 are likely more prone to respiratory and cardiac motion artifacts.

FIG. 2.

Standard midsagittal T2-weighted MR image and DBSI-derived axial fractional anisotropy maps from a CSM patient before and after C4–6 ACDF surgery. The C3, C4, C5, and C6 spinal cord segments were assessed with 7.5-mm slices at the midvertebral level. Preoperatively, the C3 level produced the best resolution compared with the C4, C5, and C6 levels. Postoperatively, image quality at the C3 level remained intact, whereas metallic artifact caused significant signal distortions in the instrumented levels (C4–6).

For these reasons, we ultimately elected to extract DBSI data from the rostral spinal cord at the C3 level for both CSM patients and healthy controls to maintain homogeneity among patient data during analyses. Although the selection of a single spinal segment for analysis has been performed in previous studies,⁸ alternatively, we could have individualized extraction protocols based on viable spinal segments. For example, if the patient underwent a C6–7 ACDF, we could theoretically extract data from C3, C4, and C5. However, this would lead to significant heterogeneity among our data, and in our preliminary analysis, we found that extracting data from the C3 level alone promoted the most homogeneity. Furthermore, as performed previously,¹⁵ we could have focused on extracting metrics from the level of maximum cord compression (LMC). Although viable on baseline evaluations, the LMCs were often instrumented segments, which were unable to be assessed on postoperative dMRI evaluations because of metal artifacts.

One potential shortcoming in focusing on the rostral C3 level is that the majority of CSM pathology is caudal to this segment, namely at the C4 to C7 levels because of the increased motion tolerable at these levels.^21,22 In CSM, chronic spinal cord compression triggers inflammatory and vascular changes, which ultimately result in neuronal apoptosis.²³ Secondary neurodegenerative changes involve axonal degeneration and accompanying demyelination at the site of compression²⁴ and classically, Wallerian, or anterograde degeneration of axons from the site of compression (i.e., spreading caudally).²⁵ From a pathophysiological standpoint, we were concerned that the C3 level was too proximal from the site of injury and would simply reveal normal, intact spinal cord microstructure.

However, both ex vivo animal studies and in vivo spinal cord imaging evidence have documented retrograde degeneration of axons (i.e., toward the proximal cell body) and transsynaptic changes within the gray matter that spread cranially.^26,27 Although these pathologies are often difficult to quantify on conventional T2-weighted imaging, recent work has reported evidence of microstructural tissue damage above the level of the lesion.^28,29 In a series of postoperative CSM patients, Ma et al. reported significant decreases in diffusion tensor imaging and neurite orientation dispersion and density imaging parameters at the nonstenotic rostral C2 level.⁸ These results suggest that microstructural changes remote from the core lesion can also be affected.

Feasibility and Efficacy of the C3 Segment to Characterize CSM Pathology

Baseline Measurements

Consistent with previous studies, the DBSI fiber fraction, which reflects axonal density, was significantly lower in CSM patients.³⁰ However, DBSI fractional anisotropy was not significantly different between groups (p = 0.31). Important to note is that, opposed to the previously described diffusion tensor imaging–derived fractional anisotropy,³¹ which averages fibers with surrounding structure properties in a voxel, DBSI fractional anisotropy is more fiber specific and is thought to assess the integrity of residual axons with limited extra-fiber confounders.³⁰ Comparable DBSI fractional anisotropy between groups suggests that residual fibers in CSM patients were not different from those of healthy controls.

Given the known pathophysiology of CSM, higher values of radial diffusivity, a measure of demyelination, were expected in CSM patients (p = 0.03). Interestingly, however, DBSI axial diffusivity was also paradoxically increased in CSM patients. Axial diffusivity is traditionally regarded as a measure of axonal injury,^32,33 and as such, we expected CSM patients to have significantly lower levels than healthy controls. However, as discussed previously,³⁰ this phenomenon may reflect contamination from extra-axonal confounders present in CSM patients. Indeed, the restricted fraction, which reflects inflammation and cellularity (e.g., microglial infiltration), and the nonrestricted fraction, which measures edema and tissue loss (e.g., vasogenic edema), were significantly higher in CSM patients than in healthy controls (p = 0.006 and p = 0.024, respectively).

Furthermore, we performed a sensitivity analysis comparing baseline C3-level DBSI metrics with those extracted from combined C3 to C6 spinal cord data, observing largely consistent results, except for DBSI radial diffusivity, which was no longer statistically significant when combining levels C3 to C6 (p = 0.06) (Supplemental Table 4). A similar sensitivity analysis was not performed at the 2-year follow-up because of metallic artifact in postoperative CSM patients.

Postoperative Evaluations

In the postoperative setting, we appreciated a similar significance pattern in DBSI parameters between CSM patients and healthy controls (Table 3). These results are encouraging because they are clinically intuitive, demonstrating the feasibility of using the rostral C3 level in evaluating CSM patients after surgery.

However, although our CSM cohort experienced a significant improvement in the mJOA score (+1.3 [SD 1.9] points, p < 0.001) at the 2-year follow-up, we did not appreciate many corresponding significant findings in the change in DBSI metrics. Of note, given that the mature CNS is extremely limited in its ability to regenerate axons,³⁴ we did not expect significant improvements in DBSI fractional anisotropy and fiber fraction—reflections of residual white matter tracts and axonal density, respectively—or DBSI axial diffusivity, a marker of axonal integrity. However, we did appreciate a significant decrease in DBSI radial diffusivity among CSM patients (Table 4), which likely explains the nonsignificant difference in postoperative DBSI radial diffusivity values between CSM patients and healthy controls (Table 3). Given that increasing radial diffusivity reflects demyelination, significant decreases in these values imply improvements in myelin integrity at the 2-year follow-up. With a growing body of literature describing the regeneration of oligodendrocytes after injury,^35,36 these results may suggest that decompressive surgery after CSM provides an opportunity for axons to remyelinate.

Aside from DBSI radial diffusivity, no other DBSI metric changed significantly at the 2-year follow-up. Importantly, however, no DBSI metric worsened, suggesting that cervical decompressive surgery may help halt the progression of myelopathy. These results may also suggest that neurological improvement may occur regardless of significant changes in these representations of spinal cord microstructure.⁹ In addition, given that we focused on the rostral C3 level, we may not have captured improvements in spinal cord microstructure at the operated levels that were harder to quantify. Alternatively, the neurological recovery may be secondary to improved neuronal excitability (such as improvements in motor evoked potentials) facilitated by reperfusion of white matter tracts after decompression.³⁷

Importantly, we also found that the change in DBSI metrics remained stable in healthy controls (Table 4). Although healthy controls may experience age-related degeneration of the spinal cord in the 2-year follow-up between MRI assessments, we do not expect this deterioration to be clinically or radiologically significant. Indeed, among healthy controls, there were no DBSI metrics that changed significantly between the baseline and 2-year scans, demonstrating the stability of DBSI. This finding suggests that DBSI analysis is not subject to substantial variability between measures of different time points, indicating that the significant results in our data are more likely reflections of true pathophysiology than random error.

Findings in Context

When comparing C3-level DBSI metrics with mJOA score improvement (as both a categorical and continuous variable) across all time points, we found few significant associations, most of which were limited to the preoperative period. Consistent with previous studies by our group,^30,38 DBSI markers of extra-axonal pathology were associated with worse mJOA scores and were significantly greater in CSM patients who did not improve after surgery. In the context of CSM pathophysiology (i.e., chronic spinal cord compression causing blood spinal cord barrier loosening and extravasation of inflammatory cytokines), our results highlight the potential significance of extra-axonal pathology in CSM disease pathways. In contrast, although we have shown previously that certain DBSI metrics differ significantly among CSM subtypes, we did not find similar results in this cohort.

In addition, we were surprised that DBSI radial diffusivity, a reflection of demyelination, was no longer statistically significant in our sensitivity analysis, given that it is known to be elevated in CSM patients.³⁰ In CSM, demyelination increases the rate of diffusion of water perpendicular to the axons (i.e., less of a barrier for water molecules to diffuse through).^39,40 One potential explanation for this nonsignificant finding may be the higher levels of cellularity—as reflected by restricted fraction—in the combined C3 to C6 data versus the C3 data alone (Supplemental Table 4). Specifically, increased cellularity, or inflammatory cells (namely microglia in CSM²⁶), invades the interaxonal space, filling the space between axons (Fig. 3). This increases the resistance for water molecules to diffuse perpendicularly to axons, dampening the effect of demyelination, theoretically leading to falsely lower radial diffusivity measurements.

FIG. 3.

Traditional and cross-sectional views of CSM spinal cord microstructural components. The illustrated pathology includes axonal injury, demyelination, and cellularity. To demonstrate microglial infiltration of axonal bundles, the interaxonal space is artificially widened. Cellular infiltration may increase the resistance of water molecules to diffuse perpendicularly to axons, dampening the effect of demyelination, which may theoretically cause falsely lower radial diffusivity measurements. © Justin Zhang, published with permission.

Taken together, our results demonstrate the ability of C3-level DBSI data to characterize representations of spinal cord microstructure after decompressive surgery for CSM. In addition, our findings suggest that when compared with combined C3 to C6 data, the C3 level alone was generally sufficient on baseline evaluations to differentiate spinal cord microstructure between CSM patients and healthy controls. In addition, we found that the level of surgery (including C3 vs below C3) did not affect DBSI values. These results are encouraging as they demonstrate not only the ability of postoperative DBSI to characterize spinal cord recovery but also its potential to significantly decrease scanner time requirements.

On the other hand, the majority of our findings comparing C3-level DBSI data and the mJOA score were negative (Supplemental Tables 5–7). Given the exploratory nature of our study, these nonsignificant results may be a reflection of our small study cohort (n = 27). They may also suggest that DBSI data from the C3 level alone may be limited in their ability to inform clinical status. Alternatively, as reflected by recent work from other groups,⁴¹ the mJOA score may provide an incomplete clinical assessment of CSM disease burden. Consequently, although out of the scope of this study, in future work we intend to evaluate the association of C3-level DBSI metrics with other neurofunctional assessments, quality of life, and pain-related disability.

Limitations

This study has several limitations. This was a single-center study with an institution-specific imaging protocol; therefore, our results lack generalizability. In addition, we focused on the C3 level, and therefore we did not incorporate information on white matter tracts at caudal spinal segments. Furthermore, our study did not evaluate long-term imaging data in CSM patients who did not undergo surgical intervention. Although comparing operative and nonoperative CSM cohorts would provide meaningful insights, the clinical equipoise of such a study design is lacking.

Conclusions

DBSI has emerged in recent years as a particularly robust diffusion-weighted imaging modality to assess patients with CSM. However, no study to date has evaluated its use in informing neural recovery after cervical decompressive surgery. We demonstrated that DBSI data extracted from the C3 level could be used to differentiate representations of spinal cord microstructure between CSM patients and healthy controls in both the preoperative and postoperative settings. We describe changes in C3-level DBSI metrics after surgical decompression and highlight the stability of these imaging parameters in a healthy control population. These advanced imaging metrics may have use as potential noninvasive biomarkers of neurological recovery following cervical decompressive surgery.

ABBREVIATIONS

ACDF

anterior cervical discectomy and fusion

CSM

cervical spondylotic myelopathy

DBSI

diffusion basis spectrum imaging

dMRI

diffusion-weighted MRI

MCID

minimal clinically important difference

mJOA

modified Japanese Orthopaedic Association