Segmented quantitative diffusion tensor imaging evaluation of acute traumatic cervical spinal cord injury

Abstract

Objectives:

To evaluate segmented diffusion tensor imaging (DTI) white matter tract fractional anisotropy (FA) and mean diffusivity (MD) values in acute cervical spinal cord injury (CSCI).

Methods:

15 patients with acute CSCI and 12 control subjects were prospectively recruited and underwent axial DTI as part of the spine trauma MRI. Datasets were put through a semi-automated probabilistic segmentation algorithm that analyzed white matter, motor and sensory tracts. FA and MD values were calculated for white matter, sensory (spinal lemniscal) and motor tracts (ventral/lateral corticospinal) at the level of clinical injury, levels remote from injury and in normal controls.

Results:

There were significant differences in FA between the level of injury and controls for total white matter (0.65 ± .09 vs 0.68 ± .07; p = .044), motor tracts (0.64 ± .07 vs 0.7 ± .09; p = .006), and combined motor/sensory tracts (0.63 ± .09 vs 0.69 ± .08; p = .022). In addition, there were significant FA differences between the level of injury and one level caudal to the injury for combined motor tracts (0.64 ± .07 vs 0.69 ± .05; p = .002) and combined motor/sensory tracts (0.63 ± .09 vs 0.7 ± .07; p = .011). There were no significant differences for MD between the level of injury and one level caudal to the injury or normal controls.

Conclusion:

Abnormalities in DTI metrics of DTI-segmented white matter tracts were detected at the neurological level of injury relative to normal controls and levels remote from the injury site, confirming its value in CSCI assessment.

Advances in knowledge:

Segmented DTI analysis can help identify microstructural spinal cord abnormalities in the setting of traumatic cervical spinal cord injury.

Introduction

Diffusion tensor imaging (DTI) is a technique that provides micro-structural evaluation not afforded by conventional MRI techniques. DTI provides various quantitative metrics including mean diffusivity (MD), radial diffusivity (RD), axial diffusivity (AD), and fractional anisotropy (FA). This technique has been applied to spinal cord disease and has shown the ability to detect spinal cord abnormalities in demyelinating disease, ¹ spinal cord injury, ^2,3 HIV myelopathy, ⁴ spondylotic myelopathy, ^5,6 and various inflammatory ^7,8 and vascular myelopathies. ⁹

In the setting of traumatic cervical spinal cord injury (CSCI), DTI has shown the ability to detect abnormalities in regions of normal appearing spinal cord, detect abnormalities in asymptomatic spinal cord compression, ¹⁰ significantly correlate with degree of injury better than conventional MRI, ³ indicate more severe injury in the setting of cord hemorrhage, ² predict motor outcomes ¹¹ and shows strong correlation with cord regeneration in the post-traumatic setting after stem cell therapy in animal models. ¹² DTI has also been shown to be a relatively reproducible technique in the setting of acute CSCI. ¹³

Early studies of brain DTI relied on manual regions of interest for analysis. However, DTI brain studies now utilize segmentation tools to provide more reproducible, accurate, and reliable results. ¹⁴ Similarly, most spinal cord DTI studies have used manual regions of interest. However, a probabilistic atlas-based segmentation tool, “Spinal Cord Toolbox” (SCT) has been developed that can segment the spinal cord into 30 unique spinal cord white matter tracts. ¹⁵ SCT has been used to evaluate multi-parametric MRI quantitative values in healthy volunteers ¹⁶ and has also been used to assess white matter quantitative MRI perturbations in asymptomatic spinal cord compression in degenerative spinal disease. ¹⁰ This algorithm has also been shown to be reliable and reproducible through inter-rater and test-retest analysis. ¹³ In the current study, we evaluate segmented DTI metrics in an acute traumatic CSCI population at a single institution. We compare segmented spinal cord DTI metrics in spinal cord traumatic injuries relative to normal controls.

Methods and materials

Subjects

After institutional ethics review board approval (in accordance with ethical standards from the 1964 Declaration of Helsinki and its later amendments), 50 consecutive patients presenting with acute cervical spine trauma with suspicion for CSCI between 5/2016 and 3/2017 were prospectively recruited through an institutional review board-approved waiver of consent. Consent waiver was allowed due to the acute need for MRI for these patients due to the nature of their injury and immediate need for treatment at all hours of the day and night. As part of the clinical spine trauma MRI, axial DTI of the spine was also performed. Inclusion criteria were: (1) clinical concern for acute CSCI; (2) undergoing MR of the cervical spine within 72 h of initial injury prior to intervention; (3) adult patient ≥18 years of age; and (4) evidence of acute CSCI injury on clinical evaluation with matching conventional MRI abnormality. Exclusion criteria were: (1) spine surgery or spine hardware within the scanning field; (2) pregnancy; (3) contraindication to MRI; or (4) known spinal cord disease or previous injury that would affect DTI metrics. Twelve control subjects who signed informed consent were also included in the study and underwent the same imaging sequences as the study subjects. These patients had no history of spine disease or spine symptoms and were recruited to have a similar age range as the study subjects.

MR imaging acquisition

MRI scanning was performed on a 3T Siemens Prisma scanning system (Siemens Healthineers; Erlangen, Germany). The scanning protocol included 2D sagittal T1 FLAIR, T2, STIR, axial T2, and axial DTI. For the current study, axial DTI and sagittal STIR were used in processing and analysis. Axial DTI sequences were multi-shot segmented echo-planar read-out (RESOLVE) acquired with 10 directions and six minimally weighted volumes with the following parameters: TR/TE, 2300/48 ms; in-plane resolution, 1.2×1.2 mm; slice thickness, 5 mm; field of view, 200×80 mm; slice gap, 0; averages, 2; bandwidth, 753 Hz/pixel; generalized autocallibration partially parallel acquisition, 2; b-values, 0 and 600; slices-20; coverage, 10 cm; scan time, 4 min 58 sec. Axial DTI images extended from the foramen magnum to C7-T1 vertebral body level in the craniocaudal direction. The sagittal STIR sequence had the following parameters: TR, 3700 ms; TE, 47 ms; field of view 220×220 mm; TI, 230 ms; section thickness, 3 mm; in-plane resolution, 1×0.7 mm; parallel imaging acceleration, 2; two averages; bandwidth, 252 Hz/pixel.

Analysis pipeline

A semi-automated imaging analysis pipeline was constructed, which takes an axial DTI dataset and a sagittal STIR image as input and gives vertebral level-specific DTI metrics in total gray matter and white matter, regional white matter (dorsal, lateral and ventral), and within 30 labeled white matter tracts. Single-point manual ROIs were used to assign the spinal levels, which can be completed under a minute and can be done accurately with minimal training. Further details on the processing pipeline used are available in prior publications. ^13,15 DTI metrics were evaluated in motor tracts (lateral and ventral corticospinal tracts), sensory tracts (spinal lemniscus) (Figure 1), and total white matter. DTI metrics were evaluated at the clinically determined neurological level of injury (NLI) in CSCI patients, one level cranial to the NLI as well as corresponding levels in normal controls.

Figure 1.

Axial diffusion-weighted trace map of the spine in a normal control subject with overlaid probabilistic segmentation of sensory (a) and motor (b) tracts, with corresponding axial T ₂-weighted MRI image at the same level (c). Axial diffusion-weighted trace map of a subject with spinal cord contusion with overlaid sensory (d) and motor (e) tracts, with corresponding axial T ₂-weighted MRI image at the same level (f).

Statistical analysis

The DTI metrics were summarized at the NLI in the CSCI subjects and were compared to values derived from two types of controls without injuries. One type of control was the DTI metrics at one vertebral body level above the NLI in the CSCI subjects, referred to as an internal control. The other types of control were DTI metrics measured in the non-trauma control subjects. DTI metrics at the NLI were compared to internal controls using the paired t-test. DTI metrics at the NLI were compared to the non-trauma controls using a generalized estimating equations-based regression model, adjusted by the vertebral body level, to account for non-independence among multiple observations per control, one per level. All statistical calculations were conducted with the statistical computing language R (v. 3.1.1; R Foundation for Statistical Computing, Vienna, Austria). Throughout, two-sided tests were used, with statistical significance defined as p < 0.05.

Results

Patient demographics

From the 50 spine trauma subjects enrolled, 23 were found to have clinical evidence of CSCI. From this group, 15 patients were found to have corresponding conventional MRI evidence of CSCI. Of these, 13 patients had adequate image quality and coverage of the CSCI. Demographic and injury data are listed in Table 1. The clinical injury grades were as follows: 4 AIS A, 3 AIS C, and 6 AIS D. Eight patients had follow-up ISNSCI exam, performed 11–44 days after the first exam (median of 24 days). Three patients (38%) showed improvement in clinical scores. Mean patient age for CSCI patients was 40 ± 20 years compared to 36 ± 19 for control subjects (p = 0.64).

Table 1.

Characteristics of CSCI patients and non-trauma controls

	Group
Variable	CSCI (n = 13)	Control (n = 12)	p value
Age, years	40 ± 20	36 ± 19	0.64
Male sex	12 (92.3)	10 (83.3)	0.59
Time from ED arrival to MRI, minutes	339 (39–759)	–
ASIA classification		–
A	4 (30.8)
B	0 (23.1)
C	3 (23.1)
D	6 (46.2)
NLI		–
C1	2 (15.4)
C2	1 (7.7)
C3	2 (15.4)
C4	7 (53.8)
C5	0 (0.0)
C6	1 (7.7)

NLI, Neurological level of injury.

Values are mean ± standard deviation, number (percent), or median (range).

Comparison of DTI metrics

Comparison of FA and MD values for white matter, sensory tracts, motor tracts, and combined sensory/motor tracts between the NLI, one vertebral body level above the NLI (internal control) and non-trauma controls is shown in Table 2. There were 13 trauma subjects with DTI coverage of the NLI, of which 10 also had coverage of the internal control level. The 12 non-trauma control subjects each had DTI of 4–5 levels (mean: 4.7) for comparison with the NLI of the trauma subjects.

Table 2.

DTI metrics at the neurological level of injury (NLI), at one vertebral body level above the NLI (internal control), and in non-trauma control subjects

	Group a			p value
Variable	NLI	Internal Control	Non-Trauma Control	NLI vs Internal Control b	NLI vs Non-Trauma Control c
Fractional Anisotropy
Sensory tracts (Spinal lemniscus)	0.63 ± 0.13	0.70 ± 0.09	0.68 ± 0.10	0.075	0.13
Motor tracts (Lateral/ventral corticospinal tracts)	0.64 ± 0.07	0.69 ± 0.05	0.70 ± 0.09	0.002	0.006
Combined sensory/motor tracts	0.63 ± 0.09	0.70 ± 0.07	0.69 ± 0.08	0.011	0.022
Total white matter	0.65 ± 0.09	0.67 ± 0.09	0.68 ± 0.07	0.43	0.044
Mean Diffusivity
Sensory tracts (Spinal lemniscus)	0.96 ± 0.34	1.01 ± 0.23	1.04 ± 0.20	0.27	0.84
Motor tracts (Lateral/ventral corticospinal tracts)	1.02 ± 0.24	0.99 ± 0.09	1.01 ± 0.17	0.78	0.45
Combined sensory/motor tracts	1.00 ± 0.26	1.00 ± 0.14	1.03 ± 0.15	0.70	0.71
Total white matter	1.03 ± 0.27	1.05 ± 0.19	1.09 ± 0.13	0.35	0.96

NLI = neurological level of injury.

^aNeurological level of injury (NLI) defined as the higher of the sensory and motor levels of injuries; the internal control is one vertebral body level higher than the NLI in the same trauma patients; the non-trauma subjects served as a separate control group (all levels);

^bPaired t-test;

^cWald test from generalized estimating equations model, adjusted by vertebral level, to account for multiple measurements per control.

When comparing DTI metrics at the NLI to non-trauma controls, there were significantly lower FA values at the NLI in the motor tracts (mean: 0.64 vs 0.70, p = 0.006), the combined sensory/motor tracts (mean: 0.63 vs 0.69, p = 0.022), and the overall white matter (mean: 0.65 vs 0.68, p = 0.044). When comparing DTI metrics at the NLI to the metrics at one level above the NLI, FA values were significantly lower at the NLI in the motor (mean: 0.64 vs 0.69, p = 0.002) and combined sensory/motor tracts (mean: 0.63 vs 0.70, p = 0.011). There was a trend toward lower FA values at the NLI compared to the internal control location in the sensory tracts (mean: 0.63 vs 0.70, p = 0.075). There were no significant differences in MD between the NLI and non-trauma or internal controls (Table 2).

Discussion

Diffusion tensor imaging of the spinal cord has shown the ability to predict outcomes in the setting of traumatic CSCI, ¹¹ as well as have stronger associations with clinical scores of injury than conventional MRI. ³ Diffusion imaging has also shown the ability to detect spinal cord abnormalities not otherwise seen on conventional MRI in the setting of trauma. ¹⁷

Previous studies that have evaluated spinal cord DTI in the setting of traumatic CSCI have relied on whole cord regions of interest, ^2,3,11 and this is the first study to evaluate the association of segmented spinal cord white matter tracts on DTI relative to clinical deficits. We found significant differences in DTI metrics (FA) between CSCI patients at the level of clinical injury and normal control patients for total white matter (p = .044), motor tracts (lateral and ventral corticospinal tracts; p = .006), and combined sensory and motor tracts (spinal lemniscal tracts and corticospinal tracts; p = .022). There were also significant differences in FA values between the level of injury and one level away for the lateral and ventral corticospinal tracts (p = .002), and combined motor and sensory white matter tracts (p = .011). There were no significant differences in MD between CSCI patients and normal controls, as well as between MD values between the level of injury and one level away from injury.

Current DTI evaluation approaches are either qualitative or rely on labor-intensive hand-drawn ROIs that may be prone reader-related variability/imprecision and poor reproducibility. In brain DTI, completely automated segmentation algorithms are available and have become the standard for DTI brain parcellation with application to clinical care. Recently, the “Spinal Cord Toolbox”, ¹⁵ a semi-automated probabilistic spinal cord registration, segmentation and parcellation tool, has been shown to be a reliable and reproducible tool in spinal cord DTI analysis. ¹³

Numerous studies have investigated the utility of quantitative DTI metrics to evaluate traumatic CSCI. Shanmuganathan et al ¹⁸ compared 16 patients with spine trauma, of which 10 had spinal cord contusions and six had soft tissue injuries, and eight normal controls with whole-cord ROI DTI spinal cord analysis and found ADC values significantly decreased relative to controls across all injury patient groups, and most significantly decreased in patients with hemorrhagic contusions followed by quadriplegic patients. Radial diffusivity was significantly reduced for all trauma patient groups as well, most prominently at the level of injury. In 10 youths with chronic CSCI, Mulcahey et al ³ compared DTI metrics derived from hand-drawn ROIs with conventional MRI findings relative to clinical scores of injury and found that DTI metrics had stronger correlations with International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI) scores compared to conventional MRI (p ≤ .0001). Diffusivity measures showed moderate-to-good negative relationships with 4 ISNCSCI measures (rs = -0.3 to −0.59) and FA showed moderate-to-good positive correlations with 5 ISNCSCI metrics (rs = 0.33–0.53). Cheran et al ² , in 25 CSCI patients (13 hemorrhagic and 12 non-hemorrhagic) and 11 normal controls showed significant DTI abnormalities in contused subjects compared to normal controls. For non-hemorrhagic injuries, there were strong correlations between DTI abnormalities and ASIA motor scores at the site of injury. In addition, DTI values have been shown to be abnormal at levels remote from the injury site, and correlate with the clinical completeness of CSCI and electrophysiological measures, thought to be a result of Wallerian degeneration. ¹⁹ In addition, DTI changes remote from the site of injury correlate with ISNCSCI total scores. ²⁰ In the comparison of 10 pediatric CSCI (with or without sacral sparing) patients and 25 controls, DTI metrics were compared between the three groups relative to presence or absence of sacral sparing, anal contraction, deep anal pressure and S4-5 sensation. There were significant differences in FA values between CSCI patients with sacral sparing, without sacral sparing and controls. ²¹ Different combinations of DTI metrics showed strong correlations with anal contraction (AD, RD; AUC = 0.9), deep anal pressure (FA; AUC = 0.88), and S4-5 sensation (FA, RD; AUC = 0.93). Our findings correlate with the above studies in terms of associations of DTI abnormalities at the site of injury, relative to normal controls. The primary difference is our study used a semi-automated probabilistic cord segmentation algorithm as compared to manually-drawn ROI’s for DTI evaluation.

Quantitative DTI metrics and their potential prognostic value in spinal cord injury have the potential to markedly accelerate treatment development for these diseases by use of these biomarkers to stratify randomized control trial investigation. With further confirmation of the value of DTI in prognosticating outcomes from spinal cord injury, these biomarkers could help identify those likely to respond to therapy. This group could then be randomized in future trials to evaluate differential response to new therapies relative to current standard management approaches.

There are a number of limitations to the current study. Firstly, the number of CSCI subjects included is relatively small, limiting the statistical analysis for some of the evaluated parameters. Secondly, this is a single center study at a level I trauma center, which may bias the types of injuries evaluated. The current study also excluded symptomatic spine trauma patients with normal conventional MRI. Future evaluation of DTI perturbations could help establish the limitations of conventional MRI in detecting microstructural injury.

Conclusion

Abnormalities in DTI metrics of DTI-segmented white matter tracts were detected at the neurological level of injury relative to normal controls and levels remote from the injury site. Segmented DTI analysis can help identify microstructural spinal cord abnormalities in the setting of traumatic CSCI.