The influence of conventional T₂ MRI indices in predicting who will walk outside one year after spinal cord injury

Abstract

Context/Objective: Magnetic resonance imaging (MRI) indices of spinal cord damage are predictive of future motor function after spinal cord injury (SCI): hyperintensity length, midsagittal tissue bridges, and Brain and Spinal Injury Center (BASIC) scores. Whether these indices are predictive of outdoor walking after SCI is unknown. The primary purpose was to see if these MRI indices predict the ability to walk outdoors one-year after SCI. The secondary purpose was to determine if MRI indices provide additional predictive value if initial lower extremity motor scores are available.

Design: Retrospective. Clinical T₂-weighted MRIs were used to quantify spinal cord damage. Three MRI indices were calculated: midsagittal ventral tissue bridges, hyperintensity length, BASIC scores.

Setting: Academic hospital.

Participants: 129 participants with cervical SCI.

Interventions: Inpatient rehabilitation.

Outcomes Measures: One year after SCI, participants self-reported their outdoor walking ability.

Results: Midsagittal ventral tissue bridges, hyperintensity length, and BASIC scores significantly correlated with outdoor walking ability (R = 0.34, P < 0.001; R = −0.25, P < 0.01; Rs = −0.35, P < 001, respectively). Using midsagittal ventral tissue bridges and hyperintensity length, the final adjusted R² for model 1 = 0.19. For model 2, the adjusted R² using motor scores alone = 0.81 and MRI variables were non-significant. All five participants with observable intramedullary hemorrhage reported they were unable to walk one block outdoors.

Conclusions: The MRI indices were significant predictors of outdoor walking ability, but when motor scores were available, this was the strongest predictor and neither midsagittal tissue bridges nor hyperintensity length contributed additional value. MRI indices may be a quick and convenient supplement to physical examination when motor testing is unavailable.

Introduction (Context/objective)

A spinal cord injury (SCI) causes deficits in both motor and sensory function of varying degrees depending on the severity of injury.¹ In attempt to classify spinal cord injuries under a common taxonomy, the International Standards for the Neurological Classification of Spinal Cord Injury (ISNCSCI) was created.² Findings from clinical examination (i.e. ISNCSCI testing) are commonly used to establish a person with SCI's current motor and and sensory functional status in the peri-traumatic period of SCI.^2,3 ISNCSCI scores and American Spinal Injury Association Impairment Scale (AIS) grades are physical exam measures used to predict motor function.^4,5 Clinically warranted magnetic resonance imaging (MRI) measures of spinal cord damage can also be used in tandem with physical examination to predict future improvements in neurological status and motor function after SCI.^6–11

Often following SCI, ISNCSCI motor testing may be contraindicated, especially in the case of sedation, spinal shock, concomitant lower extremity fracture, or other reasons. In these instances, quantitative spinal cord MRI may be especially useful. T₂-weighted MRI indicators of SCI include (i) hyperintensity length, (ii) midsagittal tissue bridges, and (iii) Brain and Spinal Injury Center (BASIC) scores. Spinal cord edema is observed as a signal hyperintensity within the cord ∼ three to four days after injury using T₂-weighted MRI.¹² Hyperintensity length is measured as the extent of T₂ signal hyperintensity along the spinal cord’s cranial-caudal axis, and is a predictor of future AIS grade conversion and future motor scores.^10,13 Midsagittal tissue bridges quantify the amount of spared spinal cord tissue (non-hyperintensity tissue) using a sagittal MRI,¹⁴ and this measure is predictive of neurological status and motor scores one-year after injury.^6,7 BASIC scores provide a five-point ordinal measure based on the extent of T₂ signal hyperintensity in the axial plane, and this measure is correlated with admission and discharge AIS grades.¹⁵

While AIS grades and motor scores provide a simple clinical perspective of motor output after SCI, functional walking (outside in the community) is a more complex task that requires a multitude of sensorimotor demands in order to be successful. Because the rehabilitation team may prescribe different therapies and different equipment in accordance with realistic walking goals, there is significant value in the ability to accurately predict outdoor ambulation. The utility of conventional MRI indices to predict community ambulation status one year after SCI is currently unknown.

Accordingly, the primary purpose of this retrospective study was to establish the potential of three conventional MRI indices (hyperintensity length, midsagittal tissue bridges, and BASIC scores) to predict a self-reported ability to walk outdoors in the community one year after SCI. The secondary purpose of this study was to determine if these MRI indices provide additional predictive value if initial ISNCSCI lower extremity motor scores are readily available. We hypothesize that the three MRI indices will be significant predictors,^10,14,15 but available ISNCSCI motor scores will be the strongest predictor.⁴

Methods

This was a retrospective study involving academic hospital and university research settings. The study was approved by local institutional review boards. Participant data were selected from the local SCI Model Systems Center. Inclusion criteria were: status post cervical spinal cord injury, clinical MRIs available for analyses, completed enrollment in the local SCI Model Systems between the years of 2010–2017 with one-year outcomes data available. Exclusion criteria were: concurrent traumatic brain injury beyond concussion, significant pre-existing neurological history (i.e. multiple sclerosis, cerebrovascular accident, etc).

Magnetic resonance imaging

Post-operative routine clinical T₂-weighted scans were used for MRI analyses, using a General Electric 1.5 T Signa Excite MR Scanner equipped with the 8-channel cervical-thoracic-lumbar (CTL) spine array coil. For sagittal imaging, twelve sagittal T2-weighted images of the cervical spinal cord were acquired with a two-dimensional fast relaxation fast spin echo sequence (slice thickness = 3 mm, slice spacing = 4 mm, field-of-view = 240 × 240 mm², matrix size = 256 × 256, in-plane resolution = 0.94 mm², interpolated in-plane resolution = 0.47 × 0.47 mm²). For axial imaging, thirty-two axial T2-weighted images of the cervical spinal cord were acquired with a two-dimensional fast relaxation fast spin echo sequence (slice thickness = 3 mm, slice spacing = 4 mm, field-of-view = 200 × 200 mm², matrix size = 224 × 224, in-plane resolution = 0.83 mm², interpolated in-plane resolution = 0.39 × 0.39 mm²).

Hyperintensity length, midsagittal tissue bridges, and BASIC scores were performed on all participants by researchers blinded to the clinical outcome measure, using OsiriX (Pixmeo Sarl, Geneva, Switzerland). Using the midsagittal T₂-weighted scan selected from the sagittal series, hyperintensity length was measured parallel to the spinal cord using the line tool (see Fig. 1).¹⁶ Ventral and dorsal midsagittal tissue bridges were quantified as the amount of spared tissue at the minimum distance from cerebrospinal fluid to the hyperintensity (see Fig. 1).¹⁶ BASIC scores were assigned using axial T₂-weighted scans, according to observable characteristics of the spinal cord damage from the identified axial image, selected from the series, with the most severe SCI (see Fig. 2).¹⁵ With all three indices, a high level of inter-rater reliability has been demonstrated.^15,16

Figure 1.

A representative participant’s midsagittal T₂-weighted image on the left, with a close-up view on the right. The hyperintensity is outlined, while the hyperintensity length, midsagittal ventral tissue bridge, and midsagittal dorsal tissue bridge are depicted (green, red, and yellow lines, respectively).

Figure 2.

The five BASIC scores are explained using a schematic (column 1), axial T₂-weighted MRIs of the spinal cord (column 2), and description (column 3) (figure adapted from Talbott et al., 2015).

ISNCSCI lower extremity motor scores

ISNCSI lower extremity motor scores were assessed based on the 0–5 standardized rating scale per major muscle group.² Scores from the five lower extremity muscle groups on the right and left sides were combined for a possible maximum total score of 50. ISNSCI testing has demonstrated favorable psychometric properties.¹⁷

Outdoor community walking

One year after SCI, per the SCI Model Systems Center at Craig Hospital follow-up, each participant answered yes/no to the question, “Are you able to walk (with or without mobility aid) for one street block outside?”

Statistical analysis

Data analysis was performed utilizing the statistical package for the social sciences (IBM SPSS 26.0, Chicago, IL). Descriptive statistics were summarized and assessed for potentially important differences. To examine the associations between the outcome variable of interest (self-reported ability to walk outside) and the MRI measures, the Pearson product-moment was used for continuous variables (hyperintensity length, midsagittal tissue bridges) and Spearman rho correlations for ordinal variables (i.e. BASIC scores).

Multiple logistic regression analyses were conducted to examine the most parsimonious variables to predict self-reported outdoor walking ability. Prognostic factors that were significantly correlated to continuous MRI variables (hyperintensity edema, midsagittal tissue bridges, P < .05) were eligible for entrance into model one. For model two, to determine the added value of the MRI measures, the ISNCNCI motor scores were entered into this model first, then the MRI variables. Adjusted R² (Coefficient of determination) were calculated for each of the models to examine how much variance in the dependent variable was explained by the independent variables that were entered to each of the regression models.

Results

A total of 129 participants met all criteria for analysis. One year after SCI, 43 participants answered “yes” that they were able to walk one street block outside, while 86 participants answered “no.” For the walking group, 36 self-identified as male, 6 self-identified as female, average age 50 ± 15 years old, average body-mass index 25 ± 5. For the non-walking group, 71 self-identified as male, 15 self-identified as female, average age 37 ± 16 years old, average body-mass index 25 ± 6. The average number of weeks between injury date and MRI acquisition was 4 ± 2.6 weeks. The average number of weeks between injury date and ISNCSI lower extremity motor score testing was 3 ± 2.0 weeks.

Midsagittal ventral tissue bridges, hyperintensity length, and BASIC scores were significantly correlated with future self-reported outdoor walking ability (R = 0.34, P < 0.001; R = −0.25, P < 0.01; Spearman’s Rho = −0.35, P < 001, respectively. See Table 1).

Table 1.

Descriptive statistics of MRI Indices and ISNCSCI scores.

Measure	Mean (SD)	Range
Midsagittal Ventral Tissue Bridge	0.52 mm (0.76)	0–2.94 mm
Hyperintensity Length	23.14 mm (19.03)	2.92–84.9 mm
BASIC Scores	2.12 (median = 2)	Frequency of Category:
		Zero: N = 2
		One: N = 23
		Two: N = 64
		Three: N = 38
		Four: N = 2
ISNCSCI lower extremity motor scores (all participants)	10.59 (16.09)	0–50
ISNCSCI lower extremity motor scores (walkers)	30.27 (14.37)	11–50
ISNCSCI lower extremity motor scores (non-walkers)	1.5 (4.59)	0–28

BASIC scores were not used for the logistic regression models because two out of the possible five categories only had two participants represented. For model one, in step one, midsagittal ventral tissue bridge was entered and in step two, hyperintensity length was entered (Criteria: Probability of F- to enter <.05, Probability of F to remove > .10). Including both MRI variables, the final adjusted R² for model one = 0.19. See Table 2 for model one details.

Table 2.

Stepwise logistic regression to predict outdoor ambulation at one year.

Imaging Model N = 129	Predictors	Odds Ratio (95% CI)	P-value	Adj R2
Step 1	Midsagittal ventral tissue bridge	2.56 (1.52, 4.32)	<.01	.14**
Final	Midsagittal ventral tissue bridge	2.24 (1.30, 3.84)	<.01	.19*
	Hyperintensity length	0.97 (0.95, 1.00)	.06

Dependent variable: self-reported ability to walk one block one year after injury. Abbreviations: CI, confidence interval; Adj, adjusted. *P < .05; **P < .01.

A total of 117 participants (out of the initial 129) had initial inpatient rehabilitation ISNCSCI motor scores available (at hospital admission) and these data were used for model two (99 male, 18 female, average age 41 ± 17 years old, average body-mass index 25 ± 5, 35 walkers, 82 non-walkers.) In step one, ISNCSCI motor scores were entered in the model; in step two, midsagittal ventral tissue bridge was entered; and in step three, hyperintensity length was entered (see Table 2). The adjusted R² using ISNCSCI motor scores alone = 0.81. When including both MRI variables, these two variables were non-significant (P > .05) while the final adjusted R² for model two = 0.82. See Table 3 for model two details.

Table 3.

Stepwise logistic regression to predict outdoor ambulation at one year.

Imaging and Motor Scores Model N = 117	Predictors	Odds Ratio (95% CI)	P-value	Adj R2
Step 1	ISNCSCI Score	1.28 (1.17, 1.41)	<.01	.81**
Step 2	ISNCSCI Score	1.28 (1.16, 1.41)	<.01	.81**
	Midsagittal ventral tissue bridge	1.10 (0.41, 2.93)	.85
Final	ISNCSCI Score	1.30 (1.16, 1.45)	<.01	.82**
	Midsagittal ventral tissue bridge	1.22 (.43, 3.48)	.71
	Hyperintensity length	1.02 (0.97, 1.07)	.44

Dependent variable: self-reported ability to walk one block one year after injury. Abbreviations: CI, confidence interval; Adj, adjusted. **P < .01.

Five participants had observable intramedullary hemorrhage (3.9% of our participant cohort). Of these five, none answered “yes” that they were able to walk one street block outside one year post SCI.

Discussion

Consistent with previous literature,^10,14,15 our hypothesis – the three T₂-weighted MRI measures of spinal cord damage will be significant predictors of self-reported outdoor walking ability – was supported by the current data, although the strength of associations was weak. Our research group previously found a high level of inter-rater reliability of manually-measured MRI indices.¹⁶ Also in alignment with past research,^5,18 the strongest predictor of walking ability was available ISNCSCI lower extremity motor scores. The addition of T₂-weighted MRI measures did not enhance predictive value, consistent with a recent study.⁴ Our results confirm the notion that, if available, neurological status and corresponding existing motor function at the time of hospital admission are the optimal predictors of future ambulatory status.

Hyperintensity length has been reported in past and recent literature.^8–11,19,20 One study found that postoperative T₂ signal hyperintensity length of less than 30 mm was associated with a high likelihood of AIS grade improvement.¹⁰ Our data suggest that shorter hyperintensity length may be a weak but statistically significant predictor of the ability to walk outside one year after injury. When added to midsagittal ventral tissue bridges in our model 1, step 2, this variable was still statistically significant but only marginally added to the model’s predictive value.

Recently, midsagittal tissue bridges predicted a higher AIS conversion rate one year after injury.²¹ Further, midsagittal ventral tissue bridges were found to be predictive of one-year follow-up outcomes including ISNCSCI lower extremity motor scores, the Spinal Cord Independence Measure, and characteristics of transcranial magnetic stimulation motor-evoked potentials (MEPs) recorded at the foot.⁶ In accordance with this study, the midsagittal ventral tissue bridge measures reported here were significantly predictive of outdoor walking ability one year after SCI. Lateral corticospinal tracts are associated with recovery of motor function following SCI.^22,23 In preclinical rat models, neuronal sprouting of the lateral corticospinal tracts (above the lesion site) into the ventral descending spinal cord motor systems was important for recovery of motor function after SCI.^24,25 Likely, our midsagittal ventral tissue bridge measure is indirectly assessing the integrity of these ventral descending spinal cord motor systems.⁶

Per a previous report using MRIs from hospital admission, BASIC scores were strongly correlated with AIS grades upon hospital discharge.¹⁵ Our data were in agreement with this finding, as a significant Spearman’s rho correlation between early imaging BASIC scores and outdoor walking ability one-year after SCI was revealed, although our strength of association was weak. It is also important to note that BASIC scores were not included in our logistic regression models because in our data set we did not have enough participants that fell into the “zero” or “four” categories (N = 2 participants for each). Using axial T₂-weighted MRI, this measure may be useful as a quick and convenient supplement to physical examination, but future research is warranted with an expanded dataset including more participants in category “zero” and category “four.”

Based on the R² values, hyperintensity length, midsagittal ventral tissue bridges, and BASIC scores explained 12%, 6%, and 12% of the variance in self-reported outdoor walking ability, respectively. Collectively, these measures demonstrated weak but statistically significant associations. Other ways of characterizing the lesion such as the ratio of damage to surrounding spinal cord,²⁶ or full volumetric measures,²⁷ may provide stronger associations with future function. However, the lesion characteristics reported here are arguably easier to measure and may be clinically useful as a first-pass assessment of injury severity.

The presence of spinal cord intramedullary hemorrhage is widely understood to be negatively associated with recovery,^{9,11–13,28–30} with only one study reporting results to the contrary.³¹ Our results support this negative association, as all five of our participants with observable hemorrhage reported that they could not walk one block outside a year after injury.

Limitations

Interestingly, for our data, age was significantly positively correlated to our ambulation outcome measure, meaning the older the age the more likely to report “yes” to the ability to walk one block outside (R = 0.38, P < 0.01). This is in contrast to two validated clinical prediction rules for future walking, both of which include younger age as a variable for a favorable prognosis.^5,18 While we are unsure of the exact reasons for our finding, it is possible that patients with SCI at our rehabilitation hospital who are younger tend to have more severe, traumatic injuries compared to those who are older (unpublished data). The mechanisms of injury were unavailable for this project and this is an acknowledged limitation.

Another limitation to note is the lack of standardized, high-resolution MRIs collected at distinct timepoints across participants. For our dataset, the imaging was collected an average of 4 ± 2.6 weeks post SCI and thus some images were outside the acute time window. Ideally, we would desire to use high resolution MRI of the cervical spinal cord (i.e. <1 mm³ voxel size) to quantify the extent and location of damage with imaging collected at the same time point across participants. Regarding potential optimal timing, midsagittal tissue bridges measured using MRI at one month post SCI were demonstrated to be predictive of future clinical outcomes.^6,7,21

In 2015, common data elements for SCI clinical research were published based on the National Institute of Neurological Disorders and Stroke (NINDS) project comprised of international working groups.³² These common data elements include recommendations from the Imaging Working Group,³³ which are important for enhancing the imaging research and understanding of SCI on a global scale. A limitation of our study is that we chose to focus on the predictive value of T₂-weighted signal hyperintensity measures of cord damage. For future research, a more robust predictive model may include the full set of NINDS common data elements, including a variety of categoric, ordinal, and continuous measures related to spinal canal and cord measurements, the length and location of edema, and spinal cord hemorrhage.³³ Using principle components analysis of a variety of variables measured within 24 h of injury, measures of intrinsic cord signal abnormality accounted for neurological impairment better than extrinsic cord compression measures.³⁴ A recent study looking at the 17 distinct NINDS common data elements found consistent good-to-excellent interrater agreement for the ordinal measures but a variable, moderate-to-low interrater agreement for the continuous measures.³⁵

Conclusion

In this retrospective study involving 129 participants with SCI, conventional T₂ MRI measures were significant predictors of self-reported outdoor walking ability. However, when ISNCSCI lower extremity motor scores were available, this was the strongest predictor and neither the midsagittal ventral tissue bridge or hyperintensity length contributed additional predictive value. Conventional T₂-weighted MRI measures may be useful as a quick and convenient supplement to physical examination, especially if motor testing is unavailable or contraindicated.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Data availability statement

Data available upon reasonable request from the authors.

Disclaimer statements

Contributors None.

Funding This work was supported by the NIH National Institute of Child Health and Human Development National Center for Medical Rehabilitation Research under Grant R03HD094577.

Conflicts of interest Authors have no conflict of interests to declare.