Utility of diffusion basis spectrum imaging in quantifying baseline disease severity and prognosis of cervical spondylotic myelopathy

Abstract

STUDY DESIGN

Prospective cohort study

OBJECTIVE

To assess the association between diffusion tensor imaging (DTI) and diffusion basis spectrum imaging (DBSI) measures and cervical spondylotic myelopathy (CSM) clinical assessments at baseline and two-year follow-up.

SUMMARY OF BACKGROUND DATA

Despite advancements in diffusion-weighted imaging, few studies have examined associations between diffusion MRI markers and CSM-specific clinical domains at baseline and long-term follow-up.

METHODS

A single-center prospective cohort study enrolled fifty CSM patients who underwent surgical decompression and twenty controls from 2018-2020. At initial evaluation, all patients underwent diffusion-weighted MRI acquisition, followed by DTI and DBSI analyses. Diffusion-weighted MRI metrics assessed white matter integrity by fractional anisotropy, axial diffusivity, radial diffusivity, and fiber fraction. To improve estimations of intra-axonal anisotropic diffusion, DBSI measures intra-/extra-axonal fraction, and intra-axonal axial diffusivity. DBSI also evaluates extra-axonal isotropic diffusion by restricted and non-restricted fraction. Clinical assessments were performed at baseline and two-year follow-up and included the mJOA, SF-36 PCS, SF-36 MCS, NDI, MDI, and DASH. Pearson’s correlation coefficients were computed to compare associations between DTI/DBSI and clinical measures. A False Discovery Rate correction was applied for multiple comparisons testing.

RESULTS

At baseline presentation, of 36 correlations analyzed between DTI metrics and CSM clinical measures, only DTI fractional anisotropy showed a positive correlation with SF-36 PCS (r=0.36, p=0.02). In comparison, there were 30/81 (37%) significant correlations among DBSI and clinical measures. Increased DBSI axial diffusivity, intra-axonal axial diffusivity, intra-axonal fraction, restricted fraction, and extra-axonal anisotropic fraction were associated with worse clinical presentation (decreased mJOA, SF-36 PCS/MCS, and increased NDI, MDI, DASH). At latest follow-up, increased preoperative DBSI intra-axonal axial diffusivity and extra-axonal anisotropic fraction were significantly correlated with improved mJOA.

CONCLUSIONS

Our findings demonstrate that DBSI measures may reflect baseline disease burden and long-term prognosis of CSM as compared to DTI. With further validation, DBSI may serve as a non-invasive biomarker following decompressive surgery.

Level of Evidence:

III

1. INTRODUCTION

Cervical spondylotic myelopathy (CSM) is the leading cause of progressive disability in patients over 65 years of age,¹ with an estimated annual incidence of 41 per million in North America.² Despite increasing knowledge, the complex pathophysiology, heterogeneous presentation, and widely variable treatment response in CSM make management decisions and outcome prognostication particularly difficult.²

Quantifiable microstructural diffusion-weighted imaging modalities, such as diffusion tensor imaging (DTI), have emerged in recent years to combat these limitations, with the goal of serving as potential biomarkers for CSM.³ Recently, our lab developed diffusion basis spectrum imaging (DBSI),⁴ designed to model and differentiate effects of coexisting inflammation, demyelination, and axonal injury that often confound DTI analyses.^{5, 6} Despite these advancements, associations between DTI and DBSI markers and CSM-specific clinical domains in the pre-and post-operative setting remain unknown. In addition, existing studies are limited by small sample sizes, sparse clinical measures, and an absence of long-term follow-up.⁷

An improved understanding of the application of DTI and DBSI in CSM may provide greater context into disease severity and offer guidance on management strategies. Therefore, we assess the relationship between DTI/DBSI measures and key CSM clinical assessments at initial presentation and two-years following surgical decompression in a single-center cohort of CSM patients.

2. METHODS

2.1. Study Design and Patient Population

A prospective cohort study enrolled fifty adult CSM and twenty healthy control patients between 2018 and 2020. This study was designed and reported according to STROBE guidelines⁸ and approved by our institution’s IRB. Informed consent was obtained from all patients. The inclusion criteria included a history of ongoing spinal cord compression and clinical evidence of CSM determined by an attending neurosurgeon. Patients with concomitant thoracic and/or lumbar stenosis or coexisting pathologies such as multiple sclerosis or rheumatoid arthritis were excluded. All CSM patients underwent decompressive cervical surgery, with the treatment selected according to the surgeon’s discretion. Anterior surgery included ACDF and arthroplasty, whereas posterior surgery included posterior cervical fusion and laminoplasty. Both CSM and control cohorts were followed longitudinally, for up to two years.

2.2. Clinical Assessments

Information on demographics, clinical characteristics, and surgical procedures were collected. At the initial preoperative assessment, complete neurological examinations, symptom duration, and hand grip dynamometer readings were obtained. Neuromuscular function was assessed via the mJOA,¹ myelopathy disability index (MDI),⁹ and disability of the arm, shoulder, and hand (DASH).¹⁰ A lower score for the mJOA and higher score for MDI and DASH reflect worse neurofunctional status. Hand grip dynamometer readings were measured to objectively assess hand grip strength. Quality-of-life was evaluated using the 36-Item Short Form Survey (SF-36) physical component summary (PCS) and mental component summary (MCS) scores,¹¹ where greater values indicate higher quality-of-life. Pain assessments were measured using the neck disability index (NDI),¹² with higher scores denoting greater pain. As such, lower scores on the mJOA, SF-36 PCS, and SF-36 MCS versus higher scores on the NDI, MDI, and DASH were considered worse clinical status.

2.3. Diffusion Basis Spectrum Imaging

All MRI data were acquired using a 3T Prisma scanner with vendor-supplied sequences. DTI assumes that diffusion-weighted MRI (dMRI) signals can be represented by a single tensor within image voxels. Conversely, DBSI models dMRI signals as a combination of distinct anisotropic tensors, reflecting axonal diffusion properties, and isotropic tensors, describing extra-axonal diffusion (including both the inside and outside of nearby cells). DBSI-derived anisotropic metrics include fractional anisotropy³ (i.e., overall white matter tract integrity) and fiber fraction (reflecting axonal density).¹³ Axial diffusivity and radial diffusivity represents diffusion of water parallel and perpendicular to axons, respectively, and are thought to assess axonal and myelin integrity, respectively.⁴ DBSI-derived isotropic tensors include restricted fraction, reflecting cellularity, and non-restricted fraction, representing tissue disintegration, edema, or contaminating cerebrospinal fluid (CSF).⁵ In addition, DBSI models intra-axonal dMRI signals. Intra-axonal axial diffusivity minimizes extra-axonal water signal contributions, offering greater sensitivity for axonal injury. Intra-axonal fraction thus also reduces extra-axonal water confounds. Extra-axonal (anisotropic) fraction is a surrogate for vasogenic edema extent that is at close vicinity of axonal fibers.

An in-lab Python-implemented pipeline was used to process all DTI and DBSI data of the spinal cord. Manually drawn regions of interest (ROI) were applied to extract DBSI metrics in the lateral, dorsal, and ventral white matter regions. Please refer to the Methods, Supplemental Digital Content 1, for detailed information behind dMRI data acquisition, DBSI modeling, and metric derivation.

2.4. Statistical Analysis

Descriptive data were analyzed via univariate analyses. Pearson’s correlation was utilized to examine relationships between radiological and clinical measures. Following the Evans’ empirical classifications of interpreting correlation strength,¹⁴ correlations <0.20 were considered “very weak”, 0.20-0.39 “weak”, 0.40-0.59 “moderate”, 0.60-0.79 “strong”, and ≥0.80 “very strong”. A False Discovery Rate correction was applied for multiple comparisons testing, with statistical significance set a priori at p<0.05. All statistical analyses were performed in R, version 4.1.4.

3. RESULTS

3.1. Demographic, Clinical, and Operative Characteristics

Fifty CSM and twenty healthy control patients were enrolled in this study. Three patients found not to meet inclusion criteria and five CSM patients having poor quality MRI data were excluded from all analyses, yielding 23 (55%) mild (mJOA 15-17), 9 (24%) moderate (mJOA 12-14), and 10 (21%) severe CSM (mJOA <11) patients. There were no demographic differences between CSM and control patients. However, baseline mJOA and patient-reported outcome measures were worse in CSM patients (Table 1, all p<0.001). Specifically, CSM patients had lower mJOA and higher MDI and DASH scores, reflecting worse neurofunctional status. In addition, CSM patients scored lower on the SF-36 PCS and SF-36 MCS, corresponding to worse quality-of-life. The CSM cohort also possessed higher scores on the NDI, demonstrating greater pain.

Table 1.

Demographics and clinical characteristics between cohorts

Characteristic	HC N = 20	CSM N = 42	p-value
Age, years	57.8 (7.7)	56.4 (8.6)	0.52
BMI, kg/m2	27.5 (6.7)	28.5 (5.7)	0.57
Sex, M:F	10:10	26:16	0.54
Caucasian race	15 (75%)	34 (81%)	0.10
Current smoker	8 (40%)	19 (45%)	0.64
mJOA score			NA
control	20
mild (15-17)	-	23 (55%)
moderate (12-14)	-	9 (24%)
severe (0-11)	-	10 (21%)
Patient-Reported Outcome Measures
SF-36 PCS	53.8 (6.9)	38.2 (10.8)	<0.001
SF-36 MCS	56.3 (6.9)	47.4 (13.1)	<0.001
NDI	3.5 (5.2)	20.1 (11.1)	<0.001
MDI	0.1 (0.5)	5.4 (5.3)	<0.001
DASH	3.7 (6.1)	36.3 (25)	<0.001

HC: healthy control; CSM: cervical spondylotic myelopathy; mJOA: modified Japanese Orthopedic Association; SF-36: 36-item short-form; PCS: physical component summary; MCS: mental component summary; NDI: neck disability index; MDI: myelopathy disability index; DASH: disabilities of the arm, shoulder, and hand

Within the CSM cohort, median symptom duration was 13 (range 2-240) months, and 13 (31%) patients had an ASA score > 2 (Table 2). A positive Babinski’s and Hoffmann’s reflex was found in 8 (19%) and 22 (52%) patients, respectively. Twenty-three (55%) patients underwent anterior surgery compared to 19 (45%) posterior operations, with most surgeries being multilevel (33, 81%). CSM patients were followed up for 22.1 ± 7.3 months, with significant improvements in the mJOA and all patient-reported outcome measures - except the SF-36 MCS - at latest follow-up (Table 2).

Table 2.

CSM-specific clinical domains and PRO measures

Characteristic	CSM
	N = 42
ASA Score, no.
≤2	29 (69%)
>2	13 (31%)
Symptom Duration, mo.	13 (2-240)
Left Hand Dynamometry, kW	62.8 (26.2)
Right Hand Dynamometry, kW	64.4 (26.7)
Positive Babinski reflex, no.	8 (19%)
Positive Hoffman reflex, no.	22 (52%)
Surgery Type
Anterior	23 (55%)
Posterior	19 (45%)
Levels Treated
Single	9 (19%)
Multilevel	33 (81%)
Follow Up, mo.	22.1 (7.3)
Preop mJOA	14.1 (3.1)
Postop mJOA	15.2 (2.9)
Change in mJOA	1.1 (2.3) *
Postoperative PRO measures
SF-36 PCS	42.9 (11.6)
Change in SF-36 PCS	4.8 (8.5) *
SF-36 MCS	48.6 (13.8)
Change in SF-36 MCS	1.4 (10.1)
NDI	12.8 (12.4)
Change in NDI	−7.2 (6.3) *
MDI	3.8 (5)
Change in MDI	−1.6 (2.8) *
DASH	24.7 (25.5)
Change in DASH	−11.6 (13) *

^*Bolded asterisks indicate significant difference on pairwise t-test.

PRO: patient-reported outcome; ASA: American Society of Anesthesiology; mJOA: modified Japanese Orthopedic Association; SF-36: 36-item short-form; PCS: physical component summary; MCS: mental component summary; NDI: neck disability index; MDI: myelopathy disability index; DASH: disabilities of the arm, shoulder, and hand

3.2. Baseline Correlations

When associating DTI and DBSI measurements with baseline clinical assessments, all patients - CSM and control subjects - were included. Complete correlation analyses comparing DTI/DBSI metrics with clinical assessments controlled for multiple comparisons are documented in Table, Supplemental Digital Content 2. A correlation matrix depicting significant associations is shown in Figure 1. At baseline, among 36 correlations analyzed between DTI metrics and CSM clinical measures, only DTI fractional anisotropy and SF-36 PCS were positively correlated (r=0.36, p=0.02, Fig. 1, red box). In comparison, there were 30/81 (37%) significant correlations between DBSI and preoperative clinical measures. The weakest correlation was between DBSI fiber fraction and mJOA (r=0.32, p=0.04) and the greatest correlation was a very strong negative association between DBSI extra-axonal anisotropic fraction and mJOA (r=−0.83, p<0.001, Fig. 1, pink arrows). Increased DBSI axial diffusivity, intra-axonal axial diffusivity, intra-axonal fraction, restricted fraction, and extra-axonal anisotropic fraction were associated with worse clinical presentation (i.e., lower mJOA, SF-36 PCS/MCS, hand grip strength, and higher NDI, MDI, DASH). By comparison, increased DBSI fiber fraction was weakly correlated with better clinical presentation (mJOA: r=0.32, p=0.04; SF-36 PCS: r=0.34, p=0.03).

Figure 1.

Correlation matrix between cervical spondylotic myelopathy (CSM) clinical measures (y-axis) and DTI/DBSI parameters (x-axis). Significant correlations are shown, with larger and darker-colored circles representing greater magnitude/strength of the correlation. A False Discovery Rate correction was applied for multiple comparisons testing, with statistical significance set a priori at p<0.05. The dotted vertical line separates DTI (left) from DBSI (right) metrics. The red box indicates the one significant correlation in DTI analyses (DTI FA vs. SF-36 PCS; r=0.36, p=0.02). The pink arrows describe the weakest and strongest DBSI correlations (DBSI FF and mJOA: r=0.32, p=0.04; and DBSI extra-axonal anisotropic fiber and mJOA: r=−0.83, p<0.001). aAD: intra-axonal axial diffusivity; AD: axial diffusivity; ADC: apparent diffusion coefficient; AF: axonal fraction; DASH: disability of the arm shoulder and hand; DBSI: diffusion basis spectrum imaging; DTI: diffusion tensor imaging; EF: extra-axonal anisotropic fraction; FA: fractional anisotropy; FF: fiber fraction; LHD: left hand grip dynamometer; MCS: mental component summary; MDI: myelopathy disability index; mJOA: modified Japanese Orthopedic Association; NDI: neck disability index; NRF: non-restricted fraction; PCS: physical component summary; RD: radial diffusivity; RF: restricted fraction; RHD: right hand grip dynamometer; SD: symptom duration; SF-36: 36-item short form survey

3.3. Outcomes Correlations

All CSM patients were included when comparing preoperative radiological parameters with change in clinical assessments. There were no significant correlations between DTI metrics and treatment outcomes after controlling for multiple comparisons testing. However, increased preoperative DBSI intra-axonal axial diffusivity and extra-axonal anisotropic fraction revealed a weak positive correlation with improved neurofunctional outcomes, namely change in mJOA (r=0.37, p=0.02; r=0.34, p=0.03, respectively).

4. DISCUSSION

To our knowledge, few studies have explored the relationship of directional or isotopic diffusion components with mJOA severity,^{15, 16} and no study has comprehensively compared DTI and/or DBSI metrics with multiple patient-reported outcome measures in CSM. Taken together, these findings highlight the unmet need for a robust analysis into the application of DTI and DBSI in CSM. Therefore, we examined the association of these measures with baseline presentation and treatment outcomes to better understand the role these advanced imaging techniques may serve in CSM management.

4.1. Utility of DBSI in CSM

In our cohort, when comparing DTI and DBSI metrics with CSM-specific clinical measures, we found numerous significant relationships, even after controlling for multiple comparisons testing. For DTI metrics, the only significant correlation was between fractional anisotropy and SF-36 PCS (r=0.36, p=0.02, Fig. 1, red box). Given the proposed utility of DTI fractional anisotropy as a biomarker for CSM,^{3, 17} we were surprised to find few significant correlations, highlighting the limitations of relying on a singular measure as a radiological biomarker.

When compared to DTI, there was an increase in the number and magnitude/strength of significant correlations of DBSI with clinical metrics (Fig. 1). Within the spinal cord, axial and radial diffusivity have been suggested as markers of axonal and myelin integrity, with reduced axial diffusivity reflecting axonal injury and increased radial diffusivity corresponding to demyelination.^{16, 18} Therefore, we were surprised that increased axial diffusivity correlated with decreased neuromuscular function (lower mJOA, higher MDI, and higher DASH), worse quality-of-life (reduced SF-36 PCS/MCS), and increased pain (higher NDI). We hypothesize that this is due to white matter tract disintegration in CSM causing tissue loss, vasogenic edema, and CSF contamination, all of which confound DBSI modeling and inflate estimations of axial diffusivity.

In order to further interpret the relationship between directional diffusivities (i.e., axial and radial diffusivity) and clinical measures, a deeper understanding of CSM pathophysiology is necessary. Traditionally, impaired spinal cord perfusion in the setting of chronic spinal cord compression is thought to mediate CSM pathology.^{19, 20} Local ischemia is hypothesized to precipitate endothelial cell and spinal cord parenchyma dysfunction, compromising the blood-spinal cord barrier and promoting extravasation of pro-inflammatory cytokines²¹ from the peripheral circulation. This in turn contributes to increased cellularity^22–24 and vasogenic edema,²⁵ which can ultimately contribute to neuronal apoptosis.^{26, 27 28, 29} Current imaging modalities, however, are unable to quantify extra-axonal pathologies such as vasogenic edema or cellularity. A distinct advantage of DBSI, therefore, is its ability to separate anisotropic and isotropic diffusion components, providing quantitative measures such as restricted fraction, a marker of cellularity, and non-restricted fraction and extra-axonal anisotropic fraction, reflections of vasogenic edema. These quantifiable diffusion components provide additional insights into disease severity, with increasing restricted fraction and extra-axonal anisotropic fraction correlating with worse clinical presentation (decreased mJOA, SF-36 PCS/MCS, and hand grip strength, and increased NDI, MDI, and DASH) (Fig. 1).

With the effects of extra-axonal pathology in mind, we assumed that vasogenic edema was confounding DBSI assessments. Despite its efficacy, DBSI is still imperfect in accounting for extra-axonal water signals mimicking intra-axonal diffusion. Consequently, we hypothesized that markedly increased extra-axonal water signals resulting from axonal injury, vasogenic edema, and tissue loss caused increased CSF contamination, overestimating axial diffusivity values. Therefore, intra-axonal axial diffusivity was developed to be more sensitive to axonal injury by minimizing the effect of increased extra-axonal water signals present with greater CSM severity. However, when evaluating intra-axonal axial diffusivity, the direction of correlation persisted, suggesting that extra-axonal water signals (i.e., vasogenic edema) continues to confound these measurements (Fig. 1).

Of note, the persistently elevated increases in intra-axonal axial diffusivity may also be partially attributable to increases in intra-axonal water signals and not solely from confounding extra-axonal water. In order to investigate this, we evaluated intra-axonal fraction, which measures the fraction of anisotropic diffusion originating within the axon, providing context on intra-axonal water signals. We found intra-axonal fraction to be significantly elevated in CSM patients compared to controls, suggesting that CSM pathology may potentially be associated with increased intra-axonal water (see Figure, Supplemental Digital 3, which demonstrates significantly higher DBSI intra-axonal fraction measurements in CSM compared to control patients). Given previous reports of axonal swelling in spinal cord injury animal models,^33,36 the increase in intra-axonal fraction may partially reflect axonal swelling secondary to cytotoxic edema (Fig. 2). However, we contend that vasogenic edema is the dominant factor confounding the estimate of intra-axonal axial diffusivity, given the well-known increase in extra-axonal water (i.e., vasogenic edema) present in CSM.¹ Regardless, further basic research and application of DBSI is necessary to provide greater context into CSM pathobiology.

Figure 2.

Axonal bundle depicting cervical spondylotic myelopathy (CSM) spinal cord white tract microstructural components and pathology. The inter-axonal space is exaggerated to demonstrate widening in the presence of extra-axonal cell infiltration and vasogenic edema. The zoomed-in cube represents an image voxel from which diffusion-weighted MRI signals originate. In CSM, increased intra-axonal fraction may represent increased intra-axonal water signals due to axonal swelling secondary to cytotoxic edema.

4.2. Findings in Context

Although diffusion-weighted imaging techniques are increasingly utilized, there is a paucity of literature assessing their utility to improve CSM management. The vast majority of studies investigating this topic correlate DTI fractional anisotropy and mJOA severity,³ suggesting that DTI fractional anisotropy may serve as a biomarker in CSM.^{3, 17} These studies provide valuable insights; however, the pathological complexity of CSM cannot be fully described by a non-specific imaging marker like fractional anisotropy, or evaluated by one clinical measure like the mJOA. Though DTI fractional anisotropy provides a useful overall assessment of white matter tract integrity, it offers no insight on directional or isotropic diffusion components that inform CSM pathological nuances. The mJOA provides valuable insights on a patient’s global function but is subject to ceiling effects.³⁰

Our results highlight the increased granularity afforded with DBSI. Not only were DBSI correlations greater in number and magnitude but were able to capture significant, albeit weak, correlations with mental health assessments (i.e., SF-36 MCS) and objective clinical measures (i.e., hand dynamometry) (Fig. 1). In addition, we found that when assessing treatment outcomes at two-year follow-up, only increased DBSI intra-axonal axial diffusivity and extra-axonal anisotropic fraction were associated with greater improvements in neuromuscular function, namely change in mJOA. These results are consistent with the general idea posited in the CSM literature, that patients with more severe baseline disease often receive the greatest benefit after surgery.³¹ Last, we were surprised at the lack of significant associations between DTI/DBSI metrics and symptom duration, as it has been proposed to be a predictor of functional recovery.^{31, 32} This may be due to our limited sample size when compared to large multicenter databases^{33, 34} or the heterogeneity seen in symptom duration in our cohort (range 2-240 months).

Taken together, we do not suggest that clinicians should rely on a singular DBSI metric. Instead, multiple DBSI metrics taken in conjunction with clinical measures may help guide surgical decision-making. Although not in the scope of the present study, future work should incorporate diffusion-weighted imaging measures in predictive modeling approaches to assess their utility in predicting treatment outcomes, dependent and independent of clinical assessments.

4.3. Limitations

This study is limited by its single-center nature, and therefore, multicenter studies utilizing similar imaging protocols should be performed. In addition, due to anatomical complexity, robust MRI evaluations in the cervical spinal cord remain difficult. Although an institutional-specific cervical spine diffusion-weighted imaging protocol³⁵ was utilized, our findings should be interpreted in this context. Additionally, refinement of DBSI and exploration of even more advanced imaging modalities are necessary to yield more accurate assessments of CSM pathology. Nonetheless, the present study is the first to comprehensively correlate DTI and DBSI metrics with a wide range of clinical measures with a large cohort of CSM patients in the pre- and postoperative setting.

5. CONCLUSION

Quantifiable microstructural imaging modalities such as DBSI provide highly specific assessments of CSM pathology that may improve management strategies. In our cohort, we found DBSI to correlate with multiple CSM-specific domains, including neuromuscular function, quality-of-life, and pain measures, preoperatively and at long-term follow-up. DBSI may have promise as an objective, non-invasive marker for both preoperative CSM disease severity and postoperative outcome.

Supplementary Material

Supplemental Digital Content 1, Methods

Supplemental Digital Content 1, Methods.

Supplemental Digital Content 2, Table

Supplemental Digital Content 2, Table. Complete correlation analyses between DTI/DBSI metrics and clinical measures

Supplemental Digital Content 3, Figure

Supplemental Digital Content 3, Figure. Boxplot comparing DBSI intra-axonal fraction among patient groups (control, mild, moderate, and severe myelopathy). DBSI intra-axonal fraction was significantly different between control and CSM patients, without differences among mJOA subgroups. *p<0.05

• A major barrier in optimizing care for patients with cervical spondylotic myelopathy is the lack of non-invasive methods and quantifiable metrics to accurately evaluate spinal cord integrity

• Detailed evaluation of white matter tract integrity through diffusion basis spectrum imaging (DBSI) may enhance the ability of providers to treat CSM

• Our findings demonstrate significant associations between DBSI and multiple CSM clinical measures in the pre- and postoperative setting

• With further validation, DBSI may serve as a non-invasive biomarker following decompressive surgery for CSM