MRI metrics at the epicenter of spinal cord injury are correlated with the stepping process in rhesus monkeys.

Clinical evaluations of long-term outcomes in the early-stage spinal cord injury (SCI) focus on macroscopic motor performance and are limited in their prognostic precision. This study was designed to investigate the sensitivity of the magnetic resonance imaging (MRI) indexes to the data-driven gait process after SCI. Ten adult female rhesus monkeys were subjected to thoracic SCI. Kinematics-based gait examinations were performed at 1 (early stage) and 12 (chronic stage) months post-SCI. The proportion of stepping (PS) and gait stability (GS) were calculated as the outcome measures. MRI metrics, which were derived from structural imaging (spinal cord cross-sectional area, SCA) and diffusion tensor imaging (fractional anisotropy, FA; axial diffusivity, λ//), were acquired in the early stage and compared with functional outcomes by using correlation analysis and stepwise multivariable linear regression. Residual tissue SCA at the injury epicenter and residual tissue FA/remote normal-like tissue FA were correlated with the early-stage PS and GS. The extent of lesion site λ///residual tissue λ// in the early stage after SCI was correlated with the chronic-stage GS. The ratios of lesion site λ// to residual tissue λ// and early-stage GS were predictive of the improvement in the PS at follow-up. Similarly, the ratios of lesion site λ// to residual tissue λ// and early-stage PS best predicted chronic GS recovery. Our findings demonstrate the predictive power of MRI combined with the early data-driven gait indexes for long-term outcomes. Such an approach may help clinicians to predict functional recovery accurately.

Introduction

Spinal cord injury (SCI), which causes devastating neural damage, can block the transmission of sensory and motor information and thus induces serious dysfunction. Locomotion deficit, one of the common results of a thoracic SCI, severely hampers the mobility, daily life, and social interactions of patients [1,2,3]. Clinically effective treatments for SCI remain limited [4]. Accurate prediction of long-term functional outcome in the early stage of SCI is one of the goals of SCI diagnosis and can provide reasonable expectations of the degree of recovery to clinicians and patients, as well as a basis for the formulation of rehabilitation protocols [5, 6]. Although research on SCI is widespread, effective prognostic techniques for motor function improvement are lacking [7].

Magnetic resonance imaging (MRI) is a common technique used for diagnosis after SCI. Conventional MRI structural imaging provides useful anatomical information, such as injured segments, injury size, hemorrhage, and edema [8]. Diffusion tensor imaging (DTI), a special MRI technique, can detect the diffusion of water molecules in tissues noninvasively and thus has been adopted to investigate the microstructural integrity of spinal cord axons [9, 10]. Some studies have used conventional structural imaging [11, 12] or combined structural imaging with DTI [13] to identify the image predictors of long-term functional changes. Miyanji et al. [11] acquired the three quantitative indexes, maximum spinal cord compression, maximum canal compromise, and lesion length, and six qualitative indexes from MRI of 100 patients with cervical SCI to predict their American Spinal Injury Association (ASIA) motor scores at follow-up. Seif et al. [14] established a correlation between the anterior–posterior width of the cervical cord and lower extremity motor (LEM) scores by using a multiparametric MRI protocol. O’Dell et al. [15] evaluated MRI and motor functions of 10 patients with SCI and demonstrated a relationship between the midsagittal tissue bridge ratio on structural images and total distance walked in 6 min. A clinical study that combined cervical cord area and voxel-based diffusion indexes demonstrated that atrophy of the ventral horn of the spinal cord is associated with motor output, whereas the integrity of the corticospinal tracts is correlated with spinal cord independence measure (SCIM) scores [16]. In another study, Basso, Beattie, and Bresnahan (BBB) locomotor scores were successfully predicted in spinal cord-contused rats at 3 months post-SCI on the basis of spinal cord area, T2 signal intensity, and DTI indexes at day 1 and 1 month post-SCI [7]. Although these studies have established correlations between some image indicators and motor performances and identified the predictors of long-term outcomes, their predicted targets were macroscopic motor performances or scores that cannot accurately reflect the stepping process of subjects with SCI.

The spatiotemporal characteristics of gait patterns are the output of the locomotion system. Gait characteristics simplify locomotion to interpretable variables to enable the quantitative description of a motor process [17, 18]. The proportion of stepping (PS) is a useful intuitive indicator of gait ability that directly presents the success rate of the stepping process. Gait stability (GS) relies on the precise control of the temporal sequence and burst strength of lower-extremity muscle activity by the central nerve system (CNS) during walking and is an important index that reflects the consistency of the stepping process [19, 20]. GS can be extracted from the degree of variability in the limb endpoint trajectory during walking and allows for the quantitative reflection of the quality of the stepping process [21, 22].

The aim of this study was to explore whether the MRI metrics obtained from conventional and diffusion imaging after SCI were correlated with the stepping process and predictive of long-term gait status in a rhesus monkey model of partial SCI. We hypothesized that MRI metrics would be good indicators of the PS and GS in the early stage after SCI, whereas early MRI metrics and gait quality could predict the long-term restoration of the PS and GS.

Materials and Methods

Animal preparation

Ten adult female rhesus monkeys (weight, 5–6 kg; age, 4–6 years) were subjected to partial SCI. Female animals were chosen because they are better coordinated and less aggressive than males and are easy to care for after injury. Zoletil 50 (5 mg kg−1) and xylazine hydrochloride (5 mg kg−1) were injected intramuscularly to anesthetize the animals, and sodium pentobarbital (20 mg kg−1, i.v. gtt) was used to maintain anesthesia. The surgery was performed as described previously [23, 24]. Briefly, the right side of the 9th thoracic cord was injured by using microscissors after laminectomy. A piece of tissue with a length of 8–9 mm (rostral–caudal direction) and width of 2–3 mm (left–right direction) was removed under a surgical microscope. After surgery, benzylpenicillin (240 mg) and pentazocine (2 mg kg−1) were intramuscularly administered daily to all animals for approximately 5 days. MRI and kinematics-based gait examinations were performed before and 1 month (early stage) after surgery. The gait examinations were also conducted at 12 months post-SCI (chronic stage) to assess long-term functional recovery (Fig. 1). All experimental procedures were approved by the Biological and Medical Ethics Committee of Beihang University (approval number: BM20180046).

Fig. 1.

Conceptual diagram of the experimental design and analysis methods. SCI, spinal cord injury; MRI, magnetic resonance imaging.

MRI examination

MRI datasets were collected by using Siemens 3T clinical MR equipment (MAGNETOM Skyra, Siemens, Munich, Germany). Animals were positioned in the supine position, and a 32-channel receiver array spine coil was placed at the back of the body [23].

Anatomical images of the spinal cord were acquired by using a proton density (PD)-weighted sequence with the following scan parameters: TR, 3050 ms; TE, 11 ms; matrix, 320 × 320; pixel resolution, 0.6 mm × 0.6 mm; slice thickness, 2 mm; flip angle, 147°; 27 axial slices without a gap; and coverage of five cord segments (T7–11).

DTI data were acquired at the same center by using a single-shot spin-echo echo planar imaging (EPI) sequence with the following scan parameters: TR, 4500 ms; TE, 104 ms; matrix, 128 × 128; pixel resolution, 1.5 mm × 1.5 mm; slice thickness, 2 mm; b, 0 and 1,000 s/mm2; repeat time, 5; 12 noncollinear gradient directions; and 25 diffusion-weighted axial slices without a gap. Parallel acquisition was used with an acceleration factor of 3 to limit the extent of susceptibility artifacts. A twice-refocused pulse sequence was used to minimize eddy current effects [25, 26]. Saturation bands were set on the chest and abdomen to reduce physiological motion artifacts. The entire scanning time was approximately 14 min.

Gait test

The stepping process of the right (ipsilateral to the injury) hindlimbs of the animals was collected by using a Vicon system (Vicon 8, Oxford Metrics Limited Co., Yamton, UK ). A reflective marker was fixed on the second metatarsophalangeal joint, and its spatial location was recorded through multiple cameras in real time (recording frequency: 100 Hz). The animals were allowed to walk on a treadmill bipedally with their upper body restrained (speed: 0.2 m s−1), and successive stepping (>10 steps) data were captured for subsequent analysis [26,27,28,29].

Data analysis

A slice centered at the injury epicenter (including the lesion site and residual tissue) and two slices 2 cm from the damaged area (defined as remote normal-like tissue) were used for MRI analysis. Regions of interest (ROIs) were selected from the axial PD-weighted images. The lesion site, residual tissues, and remote normal-like tissues were manually extracted in accordance with the previously reported signal–intensity–variation (SIV) method [25]. In brief, SIV was calculated as SIV = (LSI − NSI) / LSI, where LSI is the averaged signal intensity within the ROI of the lesion site, and NSI is the averaged signal intensity within the residual or remote normal-like tissues. The PD-weighted SIVs were obtained by comparing the lesion site with the residual tissue (44.3% ± 1.1%) and the lesion site with the remote normal-like tissue (55.5% ± 1.6%). The thresholds of SIV between the lesion site and residual tissue (SIV ≥ 30%) and the lesion site and remote normal-like tissue (SIV ≥ 50%) were then selected as the standards for classifying borderline pixels in the residual and remote normal-like tissues to avoid partial volume effects with the lesion site [25]. The spinal cord cross-sectional areas (SCAs) of the lesion site, residual tissue, and remote normal-like tissue were obtained from the above ROIs.

DTI datasets were processed and analyzed using MedINRIA (http://www-sop.inria.fr/asclepios/software/MedINRIA). The detailed methods were reported previously [23, 25, 26]. For eddy current distortion, five b0 images in EPI datasets were averaged. All diffusion-weighted images were aligned with the averaged b0 image through line affine transformation to correct the mismatch induced by eddy current distortion between b0 and diffusion-weighted images [23]. For geometric distortion, the averaged b0 image was first registered to PD volumes using the diffeomorphic demons registration algorithm to obtain a nonrigid displacement field. Then, the deformation vector field was extracted and applied to all EPI datasets to correct for geometric distortion [23, 30].

The corrected EPI datasets were used to calculate DTI indexes. The ROIs of the lesion site, residual tissue, and remote normal-like tissue were determined on the PD-weighted images. Given that distinguishing gray matter from white matter in the injury epicenter is difficult, gray and white matters were drawn onto the ROI [25]. Three eigenvectors (v1, v2, v3) and the corresponding eigenvalues (λ1, λ2, λ3) of the diffusion tensor matrix were extracted from the above ROIs on a voxel-to-voxel basis [10, 23]. Fractional anisotropy (FA), apparent diffusion coefficient (ADC), axial diffusivity (λ// = λ1, parallel to the axonal pathways), and radial diffusivity (λ⊥ = [λ2 + λ3]/2, perpendicular to the axonal pathways) values at the lesion site, residual tissue, and remote normal-like tissue were then acquired [8]. The values of the DTI indexes at the lesion site of each animal were divided by their own residual tissue values to decrease the effect of individual variability and by their own remote normal-like tissue values (formulas are given in Fig. 2). The possible correlations among DTI indexes were detected by using Pearson correlation tests. Only noncollinear variables (FA and λ//) were retained for subsequent analysis (Fig. 2).

Fig. 2.

Correlations among DTI indexes at 1 month post-SCI. The bar indicates the correlation coefficient value. FA, fractional anisotropy; ADC, apparent diffusion coefficient; λ//, axial diffusivity (parallel to the axonal pathways); λ⊥, radial diffusivity (perpendicular to the axonal pathways). *P<0.05. **P<0.01. ***P<0.001. ****P<0.0001.

The gait datasets were processed and calculated by using custom Matlab-based (MathWorks, Natick, MA, USA) software. The gait cycle was first divided automatically [31]. Then, the step height, stride length, direction and acceleration of the limb endpoint velocity at swing onset, and path length of the limb endpoint trajectory of each gait cycle were computed [28]. The PS and GS were set to decrease the effect of individual variability and to reflect gait performance comprehensively. The PS was defined as the ratio of successful gait cycles (step height >10% of its normal value) per total gait cycle in each animal at each time point (100% indicates totally successful stepping at a time point, and 0% indicates totally dragging at a time point). For GS, each gait parameter relative to its average value for all gait cycles was first computed as the variability. The consistency of each gait parameter was then obtained with the logarithm of the variability. GS was defined as the mean value of the consistency of all gait parameters at each time point for each animal, and its value ranged from 0 (the limb endpoint trajectory was completely different) to 1 (the limb endpoint trajectory was the same). The GS value of the dragged gait cycle was set to 0.

Finally, we obtained nine MRI and gait variables in the early stage of SCI: residual tissue SCA, lesion site SCA/residual tissue SCA, residual tissue SCA/remote normal-like tissue SCA, lesion site FA/residual tissue FA, residual tissue FA/remote normal-like tissue FA, lesion site λ///residual tissue λ//, residual tissue λ///remote normal-like tissue λ//, PSearly, and GSearly. Four gait indexes in the chronic stage of SCI were also obtained: PSchronic, GSchronic, ΔPS (ΔPS = PSchronic − PSearly), and ΔGS (ΔGS = GSchronic − GSearly).

Statistics

Statistical analysis was conducted by using IBM SPSS Statistics 20.0 (IBM Corp, Armonk, NY, USA). The paired t-test was used to detect the changes in MRI and gait metrics between pre- and post-SCI time points. Bonferroni correction was applied for multiple comparisons. Pearson (normality) or Spearman (non-normality) correlation was executed to explore the relationship between MRI metrics and gait performance. The nonparametric Kolmogorov–Smirnov test was used to examine the normality of MRI and gait metrics. All MRI variables at 1 month post-SCI were included for the prediction of PSearly and GSearly. For the prediction of long-term stepping status, MRI and gait metrics at 1 month post-SCI were analyzed. Stepwise multivariable linear regression was utilized to identify the models that included only significant variables and the highest R regression [32]. The squared partial correlation (Pr) between each significant variable and gait performance was calculated to display the proportion of variability in the stepping process that could be explained by each independent variable after the effects of other variables were removed. Data are reported as the mean ± SD. The significance level was set at P<0.05.

Results

SCI alters MRI and gait variables

All monkeys were subjected to SCI and subsequent MRI and kinematic assessments. The structure of the normal spinal cord was intact, and the gray and white matters were clearly distinguishable (Fig. 3A). After injury, the spinal cord structure was damaged, and the spinal cord morphology was obviously altered. Reductions in visible tissue areas in the injured site reflected injury-induced atrophy (Fig. 3A). Longitudinal comparisons showed that the residual tissue SCA was significantly smaller than its normal states, and its ratio with the remote normal-like tissue SCA was also reduced (Fig. 3B). At 1 month post-SCI, residual tissue FA/remote normal-like tissue FA exhibited a clear decrease, whereas residual tissue λ///remote normal-like tissue λ// remained stable (Fig. 3B). Animals could complete stepping in the normal state (PS: 100%), and the stability of the hindlimb endpoint trajectory was 0.830 ± 0.032. In the early stage after SCI, the PS (52.971% ± 32.409%) and GS (0.366 ± 0.209) of the hindlimb ipsilateral to the injury exhibited pronounced deterioration (PS, Pcorrected=0.004; GS, Pcorrected=0.000). After 12 months, the stepping ability of the animals displayed significant improvement (PS, 93.571% ± 15.954%, Pcorrected=0.009 for 12 months (mo.) vs. 1 mo., Pcorrected=0.704 for 12 mo. vs. normal; GS: 0.724 ± 0.127, Pcorrected=0.003 for 12 mo. vs. 1 mo., Pcorrected=0.104 for 12 mo. vs. normal; Fig. 3C).

Fig. 3.

SCI alters the spinal cord architecture and gait performances of animals. (A) Representative axial PD-weighted structural images displaying the alterations in spinal cord morphology at the injury epicenter. (B) MRI metrics significantly changed after injury. Box plots presenting the median and 25th and 75th percentiles; whiskers indicate the minimum and maximum values. Dots represent individual values. (C) Typical results showing the successive trajectories (n>10 steps) of the hindlimb endpoint ipsilateral to the injury in the healthy state and at 1 and 12 months post-SCI. The thin arrow indicates the direction of the limb endpoint velocity at swing onset. Longitudinal changes in the PS and GS are shown. Large circles show the average per time point. Data are means ± SD. Lines represent individual variations. CSF, cerebrospinal fluid; SCA, spinal cord cross-sectional area; FA, fractional anisotropy; λ//, axial diffusivity (parallel to the axonal pathways); PS, proportion of stepping; GS, gait stability; mo., months. *P<0.05. **P<0.01. ***P<0.001.

MRI metrics reflect early-stage gait ability

The relationships of MRI metrics with the PS and GS at the same time point in the early stage after SCI were analyzed (Table 1). The ratio of residual tissue SCA to remote normal-like tissue SCA exhibited the strongest correlation with the PS (r=0.785, P=0.007) and GS (r=0.795, P=0.006), whereas the variables extracted from DTI showed no obvious correlation with either the PS or GS (P>0.05).

Table 1.

Correlations between the MRI and gait metrics at 1 month post-SCI

		Residual tissue SCA	Lesion site SCA/residual tissue SCA	Residual tissue SCA/remote tissue SCA	Lesion site FA/residual tissue FA	Residual tissue FA/remote tissue FA	Lesion site λ///residual tissue λ//	Residual tissue λ// /remote tissue λ//
PS	r	0.764*	−0.599	0.785**	−0.117	0.524	−0.158	−0.242
	P	0.01	0.067	0.007	0.747	0.12	0.662	0.5
GS	r	0.766**	−0.59	0.795**	−0.196	0.515	−0.121	−0.178
	P	0.01	0.073	0.006	0.587	0.127	0.74	0.623

MRI, magnetic resonance imaging; SCI, spinal cord injury; PS, proportion of stepping; GS, gait stability; SCA, spinal cord cross-sectional area; FA, fractional anisotropy; λ//, axial diffusivity (parallel to the axonal pathways). *P<0.05. **P<0.01.

By using stepwise multivariable linear regression, we analyzed all seven predefined variables (MRI metrics) as indicators of gait ability (PS and GS). The best model for reflecting early-stage PS (adjusted R=0.874, P=0.000) and GS (adjusted R=0.866, P=0.000) included residual tissue SCA and residual tissue FA/remote normal-like tissue FA (Table 2).

Table 2.

Multivariable regression analysis of MRI metrics as indicators of the stepping process at 1 month after injury

MRI parameters	PS				GS
	Standardized coefficient β	t value	P value	Pr2 value	Standardized coefficient β	t value	P value	Pr2 value
Residual tissue SCA	0.793	6.686	0.00028	0.865	0.795	6.518	0.00033	0.859
Residual tissue FA / remote normal-like tissue FA	0.565	4.765	0.00205	0.764	0.557	4.563	0.00259	0.748

MRI, magnetic resonance imaging; PS, proportion of stepping; GS, gait stability; SCA, spinal cord cross-sectional area; FA, fractional anisotropy; Pr, squared partial correlation.

MRI metrics are correlated with chronic-stage gait ability

MRI metrics and gait indexes in the early stage were also evaluated in terms of their relationship with the long-term stepping process. No marked correlation was observed between the PS in the chronic stage and MRI metrics at 1 month post-SCI (P>0.05). However, a close but not significant correlation between PS in the chronic stage and lesion site λ///residual tissue λ// in the early stage was observed (r=0.631, P=0.051 > 0.05; Fig. 4A). A significant relationship was found between GS in the chronic stage and lesion site λ///residual tissue λ// in the early stage (r=0.706, P=0.022; Fig. 4B). Gait indexes in the chronic stage showed no relationship with PSearly and GSearly (P>0.05). In addition, ΔPS (r=−0.635, P=0.048) and ΔGS (r=−0.642, P=0.045) were correlated with residual tissue SCA at 1 month post-SCI (Figs. 4C, D).

Fig. 4.

Relationships between PS (A) and GS (B) at 12 months post-SCI and lesion site λ///residual tissue λ// at 1 month post-SCI and between ΔPS (C) and ΔGS (D) in the chronic stage and residual tissue SCA in the early stage. PS, proportion of stepping; GS, gait stability; λ//, axial diffusivity (parallel to the axonal pathways); SCA, spinal cord cross-sectional area; r, correlation coefficient; mo., months.

By using stepwise multivariable linear regression, we analyzed all predefined early-stage variables (MRI and gait metrics) as predictors of long-term ΔPS and ΔGS. The best model for predicting the long-term ΔPS included GSearly and lesion site λ///residual tissue λ// at 1 month post-SCI (adjusted R=0.893, P=0.000; Table 3). The best model for predicting the long-term ΔGS included PSearly and lesion site λ///residual tissue λ// at 1 month post-SCI (adjusted R=0.757, P=0.003; Table 3).

Table 3.

Multivariable regression analysis of early-stage MRI and gait variables as predictors of long-term gait improvement

MRI/gait variables	ΔPS at 12 months post-SCI				MRI/gait variables	ΔGS at 12 months post-SCI
	Standardized coefficient β	t value	P value	Pr2 value		Standardized coefficient β	t value	P value	Pr2 value
Lesion site λ// / residual tissue λ//	0.376	3.416	0.0112	0.227	Lesion site λ// / residual tissue λ//	0.357	2.145	0.069117	0.397
GSearly	−0.836	−7.602	0.00013	0.777	PSearly	−0.772	−4.637	0.002378	0.754

MRI, magnetic resonance imaging; ΔPS, change in proportion of stepping; ΔGS, change in gait stability; λ//, axial diffusivity (parallel to the axonal pathways); GSearly, proportion of stepping in the early stage; PSearly, gait stability in the early stage; Pr, squared partial correlation.

Discussion

In this study, we detected the changes in MRI metrics in the early stage of thoracic SCI, as well as the changes in the long-term stepping process after thoracic SCI, and established the associations of MRI metrics with gait performances in the early and chronic stages. By using stepwise multivariable linear regression, we determined that the extents of residual tissue SCA and residual tissue FA/remote normal-like tissue FA were the key indicators of early-stage stepping ability after injury. However, only the extent of lesion site λ///residual tissue λ// at 1 month post-SCI was correlated with chronic-stage GS. For long-term gait improvements, only the extents of lesion site λ///residual tissue λ// and early-stage stepping ability showed significant correlations in multivariable regression analysis.

SCI disrupts the integrity of the spinal cord architecture. In the absence of an effective clinical regenerative technique, the residual tissue in the injury epicenter, which is the physiological basis for restoring function, is important for connecting the supraspinal center and structures below the injury level [33, 34]. Some studies have applied MRI data from acute SCI (1–3 days after injury) for predicting functional recovery [11, 13]. However, MRI signal changes are not specific to the underlying pathophysiology [35] and can reflect various processes, such as hemorrhage, cytotoxic edema, and cord swelling [36,37,38]. Since a severe pathophysiological response occurs in the acute stage after injury, the signal of residual tissue tends to be masked by those of hemorrhage and edema. For this reason, the evaluation of the signal can change considerably over time and is highly variable across patients [37], making accurate assessment of the residual tissue unfavorable. Considering the stability of MRI image quality, one month after SCI was set as the time point for MRI scans. Two clinical studies [15, 39] have shown that small residual tissues could be measured with the gradually vanishing of hemorrhage and edema at 1 month post-SCI. We retained the noncollinear FA and λ// variables to evaluate the status of spinal cord tissues but eliminated the ADC variable because it is less sensitive to the effects of SCI than other variables [8, 40]. Our findings showed a pronounced reduction in the ratio of residual tissue FA to remote normal-like tissue FA. Such a reduction was indicative of the loss of the diffusion anisotropy of water molecules in the residual tissue. However, the diffusivity parallel to the axonal pathways remained stable after injury. This showed that the axonal pathways in the residual tissue were preserved [41, 42] but underwent demyelination [43].

A previous research explored the gait performance of rhesus monkeys with C7 cord hemitransection and found a significant improvement in hindlimb dragging and an obvious recovery in the consistency of the amplitude of the hip, knee, ankle, and toe joint movements at 6 months after injury [44]. Another study on mice with thoracic cord hemitransection also reported an improvement in the length of the hindlimb gait and the myoelectric activity amplitude of the vastus lateralis (extensor) and tibialis anterior (flexor) during stepping at 3 months post-SCI [45]. The results of the present study are similar to those of previous reports in that the stepping ability of the animals was impaired in the early phase but gradually improved. This result indicated great potential for plasticity after partial SCI.

Numerous factors affect motor functions after SCI, and these factors have been analyzed in many studies [46,47,48]. Despite variations in estimates across studies, the generalized predictive outcome of ambulation and other functional activities largely depends on the severity of the injury and the scores of the related functional assessments [49, 50]. The score data used in clinical and animal SCI prediction studies are usually derived from qualitative and/or semiquantitative scales, such as SCIM scores, functional independence measure scores, ASIA impairment scale (AIS) grades, International Standards for Neurological Classification of Spinal Cord Injury classifications, spinal cord ability ruler scores, BBB scores, and Basso mouse scale scores. Although widely used, these neurological tools struggle to reflect a high degree of variability in outcomes, as well as to differentiate differences in motor processes finely, thus complicating the prediction of long-term functional changes in SCI [7, 51]. In this study, the advantages of using kinematics-based gait analysis for predictive analysis over functional assessments were (i) the availability of an objective, data-driven description of gait performances that thus avoids the subjective influence of scale scores [51, 52] and (ii) the ability to describe each stepping process and accurately distinguish gait differences, which facilitates the determination of highly variable functional outcomes.

In the early stage after SCI, the PS and GS were correlated with residual tissues in the injury epicenter; this correlation indicated the importance of residual tissue for maintaining locomotion [33, 34]. Although the ratio of residual tissue SCA to remote normal-like tissue SCA has a significant correlation with the PS and GS, its power as an indicator is weaker than that of multivariate association given that conventional structural images cannot fully reflect the status of projection pathways [35]. By contrast, the areas of residual tissue, along with the ratio of residual tissue FA to remote normal-like tissue FA, have shown strong power to reflect the PS and GS, thus suggesting that the ability to control gait relies not only on the amount of residual tissue (areas) but also the quality (microstructural integrity) of residual tissue at the injury epicenter.

Notably, no significant relationship was observed between the early- and chronic-stage gait performances in this study. In contrast to our work, a previous study established a prediction model for mobility performance in the SCIM at 1-year follow-up by using the age of the patient with SCI, the motor scores of the quadriceps femoris (L3) and gastrocsoleus (S1) muscles, and the light touch sensation of dermatomes L3 and S1 [53]. A subsequent multicenter prospective study simplified these predictors to three: age, motor score of L3, and light touch score of S1 [54]. Another clinical study on incomplete SCI combined LEM scores with pinprick scores or age-predicted walking ability in patients at 6 months post-SCI [55]. A recent study successfully predicted the recovery of independent function by using the AIS grade and LEM scores [56] but discovered that the strongest outcome predictor (LEM score) was different from that (sensory score for light touch) found in a previous study [57]. The inconsistency of the present study with previous studies may be attributed to two reasons: (i) The exclusion of sensory-related variables from our metrics may have affected our ability to reveal the connection between early and chronic gait performance. Previous studies have demonstrated that the preservation of sensory function is fundamental to the maintenance of walking ability [58, 59]. A survey by Scivoletto et al. [48] also showed that >60% of patients with light touch and pinprick preservation recovered ambulation at 1 year postinjury. (ii) Binary classification (walk or not walk) was the most commonly used outcome to reflect exercise capacity in previous studies [53,54,55, 57]. In this study, however, PS and GS were set as outcomes to reveal the improvement in the stepping process and to reflect the alterations in gait performance accurately.

According to their definitions, PS and GS reflect different aspects of gait performance. PS mainly demonstrates whether the animal achieves the swing phase, whereas GS additionally estimates the stability and consistency of the stepping process. In primates, highly stable and consistent gait cycles (GS) require more precise supraspinal control than achieving swing (PS) [27, 60, 61]. The integrity of cortical projection pathways determines the degree of supraspinal control after SCI. Consistently, a positive relationship was only found between the ratio of lesion site λ// to residual tissue λ// and GS. This result suggested that good axonal integrity in the injury zone has the potential to improve the homogeneity of gait cycles.

Although long-term ΔPS and ΔGS were associated with residual tissue SCA in the early stage of SCI, the multivariable linear regression analysis included the ratio of lesion site λ// to residual tissue λ// and early-stage gait performances as the key predictors. The standardized coefficient β of the lesion site λ///residual tissue λ// indicated positive correlations between axonal integrity at the lesion site and ΔPS or ΔGS. However, moderate changes in long-term ΔPS and ΔGS in animals with initial high GS and PS may be due to a ceiling effect [51].

This study has several limitations. First, accurate prediction results are more valuable when they are obtained early in the clinic. Our MRI datasets for the early stage, however, were acquired at 1 month post-SCI to match kinematics-based gait examinations. In general, progression of SCI pathophysiology is similar between species [62]. However, Smith et al. [63] reported that lower mammals have shorter shock phases after SCI than humans. According to this tendency, we believe that 1 month post-SCI in rhesus monkeys corresponds to a similar or even later time point in humans. Although some reports have suggested that MRI at 2–3 weeks after injury may be beneficial in assessing correlations [11, 64], prediction in the acute phase is highly useful for treatment. Second, no sensory-related variables were included in this study. Sensory-related variables should be included in further studies because the status of sensory functions after injury is important for assessing the potential for recovery. Finally, as this was a preliminary study, only the injury site of a partial transection model was investigated; this approach limited the ability of the study to reveal potential relationships between changes beyond the damaged zone and functional improvement [65, 66] and precluded it from mimicking the complexity of clinical injury types. The correlations between gait performance and widespread alterations in the CNS under various SCI types and degrees need to be explored. Furthermore, the existing evidence suggests that older SCI patients experience a similar degree of sensorimotor recovery as younger ones but with great functional deficits [67]. Since only young adult monkeys (4–6 years old) were used in this study, the applicability of the findings to older subjects still needs to be further evaluated.

Conclusions

Our study showed that MRI and kinematics-based gait examinations in the early stage after SCI were useful tools in predicting the long-term recovery of the stepping process in a primate model of partial thoracic transection. The area and microstructural integrity of residual tissue at the injury epicenter were found to be better determinants of early-stage gait performance than the ratio of residual tissue areas to remote normal-like tissues. Although diffusion anisotropy is commonly used in diagnostic studies, λ// (diffusivity parallel to the axonal pathway) may be more important than other indexes in assessing potential functional recovery because it is correlated with chronic GS and is a predictor of the prognosis of gait improvements. The increased sensitivity of the MRI and gait metrics applied in this work may allow for the quantitative evaluation of data-driven gait performances. Such an evaluation may have the potential to provide precise expectations for functional recovery in clinical studies.

Conflicts of Interest

The authors report no conflicts of interest.