Mild gait impairment (MGI) in multiple sclerosis: Using wearable technology to identify what the neurological examination cannot

Abstract

Background:

Gait impairment is a primary driver of disability in multiple sclerosis (MS). However, traditional disability measures are nonlinear and lack sensitivity to subtle changes in motor function. Objective, scalable tools for gait assessment are critically needed.

Objective:

To evaluate the agreement between a depth-sensing video system and wearable inertial measurement unit (IMU) sensors in quantifying gait metrics in people with MS (pwMS), to assess the relationship to patient-reported disability.

Methods:

Twenty-six pwMS completed a 6-meter walk wearing IMU sensors (Clario OPAL) and were simultaneously recorded with a depth-sensing camera (Microsoft Kinect). Disability was self-reported using the Patient Determined Disease Step (PDDS) scale.

Results:

Participants had a median age of 54 years (range 23 – 71), 69% were female, and the median disease duration was 65 months (0 – 382). Strong correlations with low differences were observed between devices for cadence (r>0.9, p<0.001; mean difference −5 steps/min), stride duration (r>0.9, p<0.001, mean difference 0.03-0.06 s), and gait speed (r=0.7 – 0.9, p<0.001, mean difference 0.04 – 0.15 m/s). Gait cycle phases were moderately correlated (r=0.4 – 0.7). PDDS scores correlated most strongly with speed (r= −0.61), with weaker correlations across gait parameters.

Interpretation:

Wearable and depth video systems yielded highly correlated spatiotemporal gait data obtained in a clinical setting and are more sensitive to motor abnormalities than the PDDS. These tools could potentially be widely employed for clinical gait assessment, democratizing access to sensitive qualitative measurement of disease progression in MS.

Introduction

Multiple sclerosis (MS) is a common neurologic disease, affecting approximately 300 per 100,000 individuals in the US.^1,2 It is associated with progressive disability, with most patients requiring a gait aid within 15 years of diagnosis, reducing independence, opportunity for employment, and quality of life.^3,4 Reliable and sensitive measures of motor impairment are critical to identify and track progression early in the disease course, to monitor treatment response, and to individualize rehabilitation interventions. The current gold standard for monitoring disability in people with MS (pwMS) is the Expanded Disability Status Score (EDSS)⁵, which is limited in its utility due to the time and training needed to administer the test, poor inter-rater reliability, insensitivity to small but significant changes over time, and a discontinuous qualitative scale that is heavily skewed toward assessment of severe motor disability. ^6-8 The Disease Steps (DS) scale was developed as a metric of disability in pwMS that emphasizes motor and gait impairments but can be administered by neurologists who are not MS specialists. This metric is well correlated with the EDSS.^{9 10,11} The Patient Determined Disease Steps (PDDS) assessment is a patient-reported outcome measurement that is an adaptation of the physician-reported DS scale.^11,12 The PDDS is strongly correlated with the EDSS,^13-15 and has been used broadly as a disability metric.^15-19

Research lab-based testing using specialized equipment has demonstrated important appreciable differences in gait metrics in pwMS²⁰ that correlate with level of disability and progression of MS, including walking speed, stride length, cadence and double support time.^21-23 These research studies offer important insights into gait dysfunction in pwMS. There is a growing body of evidence validating wearable inertial measurement unit (IMU) based devices for gait in MS against gold standard marker-based camera systems,^24-31 which offer the potential for efficient in-clinic assessments. Video based options using consumer-grade marker-less “3D” cameras are also of interest for remote clinical assessments and because of the potential for in home monitoring. However, limited findings are available regarding the utility of video-based gait analysis in pwMS. In this study, we evaluate the correlation between gait metrics derived from a wearable IMU-based system and an infrared depth-sensing camera system,³² and we assess the relationship of these measurements to the PDDS score to demonstrate that these systems could be integrated into clinical and remote monitoring workflow to detect important gait features in pwMS.^33-35

Methods

Participants

This study was approved by the Institutional Review Board of Mayo Clinic, Rochester, Minnesota (Institutional Review Board 08-007846), and all patients gave written consent for the use of their medical records and gait analysis for research purposes. Patients seen in the MS clinic at Mayo Clinic in Rochester, Minnesota with a diagnosis of MS were consecutively offered enrollment into the Center for MS and Autoimmune (CMSAN) biorepository for gait analysis. The CMSAN biorepository is a biobank that prospectively collects patient gait analysis, serum, spinal fluid, and clinical data. All 26 pwMS that had completed gait assessments were included.

Survey Evaluation

Patients filled out surveys using tablet devices at the time of gait assessment. Surveys included the Patient Determined Disease Steps (PDDS)^9,10,12,13 instrument and quantification of falls in the last year.

Walking assessment

Participants completed gait tasks in clinic while wearing OPAL Devices (Clario, Pennsylvania, USA) over the sternum (1 sensor), lumbar back (1 sensor), bilateral wrists (2 sensors), and dorsal surface of the feet (2 sensors) per the manufacturer guidelines for gait analysis in Mobility Lab Version 2 (Clario, Pennsylvania, USA). The OPAL device (Clario, Pennsylvania, USA) is a commercially available IMU device for measuring metrics of gait and balance validated in healthy³⁶ and neurologic populations.^36,37 Study coordinators instructed the participants to walk 6 meters, turn 180 degrees, and walk 6 meters back at their normal self- selected speed.²⁴ Two trials were performed. Gait metrics were recorded and computed in Mobility Lab software version 2 (Clario, Pennsylvania, USA) which uses data from all sensors to create a three-dimensional motion matrix to calculate gait parameters. One Azure HD Kinect (Microsoft, Washington, USA) camera was placed behind the starting point and was synchronized using iPi Recorder software (iPi Soft LLC, Lamarjean Group, Inc, Poway, CA) to record depth information related to the subject’s walking. The camera recorded normal video and depth video at 30 frames per second (FPS). Layout of the recording set up is shown in Figure 1A and was consistent between all subjects.

Figure 1:

A. Schematic of gait assessment setup. Participants walked a 6-meter path while wearing six inertial measurement unit (IMU) devices: two on the wrists, two on the ankles, one on the sternum, and one on the lumbar spine. A depth camera simultaneously recorded their movement. The light gray rectangle depicts the walking area. B. Representative still frames from the depth camera. Frames show the participant walking toward the depth camera (left to right, top to bottom). The outline color transitions from yellow to magenta as the participant approaches the camera. Data were analyzed using iPi Mocap Studio 4 Software (iPi Soft LLC, Lamarjean Group, Inc, Poway, CA), with an overlaid orange skeleton used to track body movements.

Depth video data preprocessing and extracting gait metrics.

Gait metrics were extracted from the depth video as previously described.³² Depth recordings were processed offline using the iPi Mocap Studio 4 Software (iPi Soft LLC, Lamarjean Group, Inc, Poway, CA). A skeleton was fit to the depth point cloud data to track the person walking during the assessment (Figure 1B). The turn was excluded from analyses because camera coverage was insufficient for accurate tracking at the walk apogee (optimal camera depth 0.5 – 4 m). The skeleton trajectory was smoothed using the iPiSoft jitter removal tool to remove potential outlier datapoints. Once the depth data were processed, the joint coordinates were exported using the biomech add-in to obtain absolute positions relative to the ground in the tracking software. The default skeleton available in the iPi Motion Studio provided the following marker coordinates: center of mass (com), center of mass projection to ground, hip, lower spine, middle spine, chest, neck, head, eye, effector head, clavicle (Left/Right), shoulder (L/R), forearm (L/R), hand (L/R), thigh (L/R), shin (L/R), foot (L/R), toe (L/R), effector toe (L/R). Effector terminology is defined within the iPi Motion Studio software and was not used as a parameter in the final analysis. The cartesian values of the joint coordinates were outputted in tabular format text files which were processed in python for subsequent analysis. These text files contained header information to select specific joint marker information, such as toe or com coordinates, for subsequent stride detection.

Center of mass (com) and left and right toe markers were used for detecting strides during the session. Missing frames were interpolated using the piecewise cubic hermite interpolating polynomial (pchip) algorithm.³⁸ The x-axis was aligned to the medial-lateral axis, the y-axis was aligned to the inferior-superior axis, and the z-axis was aligned to the anterior-posterior axis during gait. The com_x and com_z positions were used for subsequent gait detection and the first frame was subtracted to ensure the person was centered and moving in the positive com_z direction. The com_z and toe_z positions were used to detect strides. Strides were defined from toe-down to toe-down (t_down[n+1] – t_down[n]) of the ipsilateral leg. First, the toe_z position was subtracted from the com_z position to find the relative change in the foot from the body. Toe down was defined as the positive peak of toe_z from the com_z (find_peaks, prominence 0.5 standard deviations of the overall toe_z). Similarly, toe up was defined as the negative peak of toe_z from the com_z (in the backward direction). These events were then labeled as R_down, R_up, L_down, and L_up depending on which toe_z marker was being analyzed. This resulted in a sequence of events that was then used to detect “valid strides.” We defined a valid right stride as sequence of R_down – L_up – L_down – R_up – R_down and left stride as L_down – R_up – R_down – L_up – L_down. These events also defined the gait cycle for which subsequent metrics were calculated.

Using the detected strides, the following gait metrics were calculated:³⁹ 1) stride duration: time between toe down events (t_down[n+1] – t_down[n]); 2) stride distance: Euclidean distance between toe down events, ∣toe_x,z(t_down[n+1])– toe_x,z(t_down[n])∣ 3) step duration: time from the ipsilateral toe down to the contralateral toe down (t_{contr,down[n]} – t_ips,down[n]); 4) step distance: Euclidean distance between the ipsilateral and contralateral toe down events ∣contr,toe_x,z(t_down[n+1]) – ips,toe_x,z(t_down[n])∣. The following time based metrics were represented as a percentage of the gait cycle: 1) double support: duration both feet were on the ground 100 × ((t_contr,up[n] – t_ips,down[n]]) + (t_ips,up[n] – t_{contr,down[n]}))/(t_{ips,down[n+1]} – t_ips,down[n]); 2) swing duration: time from ipsilateral toe lifts up to touches down (t_ips,up[n] – t_{ips,down[n+1]})/(t_{ips,down[n+1]} – t_ips,down[n]); 3) stance duration: time from the contralateral toe lifts up to touches down (t_contr,up[n] – t_{contr,down[n+1]})/(t_{ips,down[n+1]} – t_ips,down[n]); 4) gait speed: stride length divided by stride time; 5) cadence: the sum of the strides for both legs divided by the sum of the stride durations calculated for the entire session, 6) maximum swing elevation: the maximal toe_y position between t_ips,up[n] and t_{ips,down[n+1]}.

Statistical Analyses

The mean and standard deviation of each metric were reported for each experimental session. Summary statistics across sessions were calculated by taking the mean of the means or mean of the standard deviations of each metric. No standard deviation was calculated for the cadence metric derived from the video data because it was calculated across the entire session.

Each gait metric was compared between the depth video and the corresponding wearable data. Spearman correlation coefficient was used to analyze the gait testing metrics (wearable, depth video) and disability survey scores (PDDS). Pearson correlations and Bland-Altman plots were calculated to determine how well the values aligned for each metric. Bland-Altman results were presented as the bias of the difference between the two modalities (wearable, depth video) along with the lower and upper limits of agreement (LOAs, 1.96 x standard deviation). ^40,41 We considered correlation coefficients >0.9 as “strongly correlated,” coefficients >=0.4 and <0.9 as “moderately correlated,” and >=0.2 and <0.4 as “weakly correlated.” A p value of 0.05 was considered significant.

Results

Gait and depth video assessments of 26 pwMS were analyzed, including 13 people with relapsing remitting MS (RRMS), 9 with secondary progressive MS (SPMS), 2 with primary progressive MS (PPMS), 1 with clinically isolated syndrome and 1 with radiographic isolated syndrome (Table 1). Gait analysis was performed at median age of 54 years (range 23-71) and 65 months from initial diagnosis (range 0-382); 69% were female (18/26). The PDDS score for the cohort was median of 1 (range 0-5), with the most severe disability in those with PPMS (table 1).

Table 1.

Demographic and clinical features of participants.

Multiple Sclerosis Diagnosis	Number of participants	Number Female (%)	Age (years) at gait analysis (Median, Range)	Time (months) since diagnosis (Median, Range)	PDDS Score (Median, Range)	Number using gait aid
Radiographic Isolated Syndrome	1	1 (100)	[40-45] *	240	0	0
Clinically Isolated Syndrome	1	1 (100)	[70-75] *	6	0	0
Relapsing Remitting MS	13	11 (85)	53 (23-64)	89 (11-382)	0 (0-2)	0
Primary progressive MS	2	0 (0)	50 (41-60)	38 (28-48)	5 (5-5)	2
Secondary Progressive MS	9	5 (56)	58 (28-66)	50 (0-382)	3 (0-5)	2
Total Cohort	26	18 (69)	54 (23-71)	65 (0-382)	1 (0-5)	4

Abbreviations: MS, Multiple Sclerosis; PDDS, Patient Determined Disease Step.

^*Single individuals identified by 5-year age epochs to avoid identification.

Correlation and accuracy of spatiotemporal gait metrics between devices

Cadence was strongly correlated between devices for both right (r=0.95, Figure 2, Table 2) and left (r=0.94) legs, with a mean difference of −5.6 steps/min and −4.6 steps/min respectively based on Bland-Altman plots (where a negative value indicates that the depth video metrics underestimate the OPAL metrics). Spatiotemporal based metrics are summarized in Figure 3 and Table 2. There was very strong correlation between devices for time-based metrics, while distance-based metrics had moderate correlation. Stride duration (r=0.97 both legs, mean difference left=0.03 s, right=0.06 s) (Figure 3) was strongly correlated between the wearable data and depth video data. Gait speed was more strongly correlated between platforms for the left leg (r=0.94), compared to the right leg (r=0.78), with small differences between the devices (mean difference 0.04 m/s left, 0.15 m/s right). Stride length measurements were correlated between platforms for the left leg (r=0.8) but not the right leg (r=0.41), with small mean differences between the two devices (0.08 m left, 0.22 m right).

Figure 2:

Cadence metrics correlation plots and Bland-Altman plots comparing wearable device and depth video. a) Correlation plot comparing cadence for the OPAL wearable device (y-axis) vs the depth video (x-axis) measured in the left leg. Black dashed line indicates y=x. Correlation, p-value, and RMSE annotated in the plot. b) Bland-Altman plot indicating the difference between OPAL wearable device and the depth video vs the average value of the two modalities for the left leg. Black dashed lines indicate the lower and upper LOAs, solid lines indicate mean difference. Values are depicted to the right of the plot. c-d) Correlation plot and Bland-Altman plot for right leg cadence metrics.

Table 2.

Summary statistics of gait metrics measured by wearable and video devices.

	Wearable		Video
	Mean of Means	Mean of Std	Mean of Means	Mean of Std
Cadence (steps/min), L	97.93	3.65	94.23	NA
Cadence (steps/min), R	98.87	5.37	93.24	NA
Stride length (m), L	0.97	0.08	1.05	0.19
Stride length (m), R	0.82	0.23	1.04	0.18
Stride duration (s), L	1.27	0.05	1.30	0.26
Stride duration (s), R	1.26	0.07	1.32	0.26
Step duration (s), L	0.63	0.03	0.66	0.21
Step duration (s), R	0.63	0.05	0.67	0.19
Gait speed (m/s), L	0.80	0.06	0.85	0.12
Gait speed (m/s), R	0.69	0.16	0.84	0.14
Swing phase (%), L	36.41	1.61	39.54	4.98
Swing phase (%), R	35.56	3.20	39.64	4.44
Stance phase (%), L	63.59	1.61	60.46	4.98
Stance phase (%), R	64.44	3.20	60.36	4.44
Double support (%), L	26.65	1.96	21.68	13.22
Double support (%), R	27.71	2.85	20.18	5.23
Elevation at midswing (cm), L	1.57	0.48	3.04	4.17
Elevation at midswing (cm), R	1.66	0.59	2.78	3.23

Abbreviations: L, left; R, right; Std, standard deviation.

Figure 3:

Correlation and Bland-Altman plots of distance and time-based gait metrics comparing wearable device and depth video. a) Stride length, b) stride duration, c) step duration, and d) gait speed for left leg and right leg. For each metric, the left-most plot depicts the Pearson correlation (r), p-value, and RMSE value, and the right-most plot depicts the Bland-Altman plot with mean difference and LOAs. Conventions similar to figure 2.

Correlation and accuracy of gait cycle metrics between devices

Phases of the gait cycle were measured as percentages of total gait cycle time. Correlation between gait cycle measurements including swing, stance, and double support time, was moderate between the two devices (range 0.4-0.7 across phases) (Figure 4). There was a small overestimation of the swing phase relative to standing phases in the depth video device compared to the wearable devices, with mean differences ranging from −7.5% to 4.1% for the different phase measurements. Elevation at mid-swing was not well-correlated (r=0.44, Figure 5), though the mean difference between devices was less than 1.5 cm on either side.

Figure 4:

Correlation and Bland-Altman plots of gait phases comparing wearable device and depth video. a) Swing phase percent, b) stance phase percent, and c) double support time for left leg and right leg. For each metric, the left-most plot depicts the Pearson correlation (r), p-value, and RMSE value, and the right-most plot depicts the Bland-Altman plot with mean difference and LOAs. Conventions similar to figure 2.

Figure 5:

Correlation and Bland-Altman plots of elevation at mid-swing comparing OPAL wearable device and depth video. a) Correlation plot for elevation at mid-swing comparing wearable device and depth video of left leg, b) Bland-Altman plot indicating the difference between OPAL wearable device and the depth video vs the average value of the two modalities for the left leg. c-d) Correlation and Bland-Altman plots of elevation at mid-swing for right leg. Conventions similar to figure 2.

Gait metric correlation with patient reported disability status

Given the correlation between the wearable device and the depth video measurements, we assessed the correlation of these metrics with the PDDS using principal component analysis (PCA) (Figure 6A) and pairwise Spearman's rank correlation analysis (Figure 6B). These analyses confirm the strong correlation for most metrics between the wearable devices and the depth video (short separation distance in the PCA) and indicate that stride duration and step duration clustered with PDDS and were strongly correlated to this metric. Of note, cadence, gait speed, stride length, and swing phase were clustered separately from the PDDS and were anti-correlated with this metric. The strongest correlations were observed for gait speed, with a Spearman coefficient >0.9 between video and wearable measurements and Spearman coefficients <−0.6 between each of these modalities and the PDDS. The correlations between devices were stronger than with the PDDS.

Figure 6.

Correlation between wearable device and depth video measurements with PDDS. (A) Principal component analysis (PCA) of gait metrics demonstrates close clustering of wearable and depth video measurements, confirming their strong agreement. Stride duration and step duration cluster with PDDS, indicating a strong correlation with disability level. In contrast, cadence, gait speed, stride length, and swing phase cluster separately and are anti-correlated with PDDS. (B) Pairwise Spearman’s rank sum correlation heat map demonstrates a three-way comparison of strength of correlation in color (range: dark red, r=−1; dark blue, r=+1) and significance in black and white (range: black, p=0, white p=1). Gait speed demonstrates the strongest correlations (>0.9 between video and wearable measurements) and strong inverse correlations with PDDS (Spearman coefficients <−0.6).

Discussion

This study demonstrates that both wearable IMUs and depth video systems can detect meaningful differences in gait within a clinical testing environment, potentially offering greater sensitivity to change than traditional stepwise disability scales. In a clinical setting, both technologies reliably replicated key spatiotemporal gait parameters. Metrics such as cadence, stride duration, and gait speed showed high correlation and minimal error across the two modalities. In contrast, measures related to gait cycle phases and mid-swing elevation exhibited lower correlation. Gait speed showed a moderate correlation with the stepwise Patient-Determined Disease Steps (PDDS) score, whereas other gait parameters were only weakly associated. In contrast, measures related to gait cycle phases and mid-swing elevation exhibited lower correlation between devices. Gait speed showed a moderate negative correlation with the stepwise Patient-Determined Disease Steps (PDDS) score. These findings highlight the potential for wearable and depth-sensing technologies to quantitatively assess disability and monitor disease progression in individuals with multiple sclerosis in the clinical setting.

The lower correlation between PDDS and gait metrics in our study was anticipated given the variability in functional status across different PDDS levels. The PDDS is a practical and easy to administer clinical tool, but its non-linear and discrete structure limits its sensitivity, particularly in capturing the full spectrum of dysfunction in patients who report normal or advanced disability. Prior studies have found weaker correlation between PDDS and EDSS in patients who have mild impairment (0-2),¹³ and more variability in measures of gait speed at PDDS of 5-6.⁴² Quantifiable gait assessments will be particularly helpful in patients with mild disability to detect and monitor subtle changes over time.

Prior work evaluating gait changes in pwMS has shown differences in velocity, stride length, step length, double support time, and swing phase duration of gait compared to healthy controls.⁴³ Consistently, these studies show smaller steps, slower speed, and less time in swing phase for MS patients relative to healthy controls. PwMS also exhibit greater variability in these spatial-temporal parameters overall compared to controls. Studies have varied extensively in methodology for assessing and measuring these metrics,^44-47 and therefore the acceptable range of clinically significant changes has not yet been established. It is important to directly compare gait analysis methodologies and clinical scales, as we have done here, to better understand what constitutes a clinically significant change in these gait metrics. Longitudinal assessment will be needed to further clarify this. Prior work has primarily measured patients in optimized gait lab settings, often with multiple trials for each test. While optimal for scientific purposes, these approaches are expensive, time-consuming, and require sophisticated devices and highly trained personnel. In contrast, the use of wearable devices – and perhaps most importantly, the use of simple depth cameras – offers a less expensive, high bandwidth approach for reliably measuring gait in patients within a clinical setting. However, these devices require appropriate validation. In the current study the data were collected by study coordinator staff using a mobile cart, with on-the-spot study enrollment and assessment in the hallway outside of the patient exam room. Notably, our findings also suggest that depth video imaging may be readily migrated from the clinic to the patient’s home, opening up a tremendous opportunity for daily longitudinal assessment of gait in a large cohort of MS patients.

We hypothesize that the differences in gait phase measurements between the two devices are primarily due to variations in detection methods: OPAL devices rely on foot-ground contact, while depth video systems track toe marker positions relative to the body’s center of mass. This pattern is consistent with a recent IMU-vs-camera validation performed by Zahn et al., which reported that foot-worn IMUs underestimate swing time and overestimate stance time relative to a marker-based video system, leading to a relative error in the swing time calculation in healthy controls that exceeded 7%.³⁰ The authors attribute this bias to differences in detection of the heel strike and toe-off events used to separate stance phases and swing phases, with the camera system using a heel-strike definition based on a force-plate threshold and the IMU defining the end of the swing phase as the transition from dorsiflexion to plantar flexion. This methodological asymmetry mirrors our findings, as the OPAL pipeline anchors phase separations to foot–ground contact, while the depth video derives toe-down/up from the toe kinematics relative to the body COM. Kinematic toe events can precede or lag true load acceptance or toe-off when light contact persists between the foot and ground, increasing swing time and reducing stance time in the video analysis. Zahn et al. also found the weakest agreement between measurement modalities for toe clearance, with the IMU underestimating the value by ~1.9 cm relative to the video measurement. Similarly, our mid-swing elevation showed the poorest cross-device agreement, with mean elevations differing by up to 1.5 cm between the devices. With further optimization and validation against a pressure-sensing gait mat, these methods have the potential for improvement. Across metrics, we observed stronger correlations for the left leg compared to the right. We suspect this is because most patients initiate walking with their right leg, requiring the depth camera to recalibrate the skeletal model after patients completed turns. This recalibration may result in inaccuracies in the first step of the gait cycle. Potential solutions include instructing patients to pause after turning or refining the algorithm to exclude the first step after a turn. Additionally, some variability may arise from patients wearing their usual clothing and footwear during testing, as well as conducting assessments in a clinical hallway with potential interruptions. However, we view these conditions as strengths, as they reflect a real-world clinical environment and demonstrate the robustness of these methods in practical settings.

This study was limited by utilizing the patient-reported PDDS, rather than rater reported EDSS. Future studies should incorporate both, particularly for individuals with low disability where there is more variability in the correlation of these metrics. In the future, important next steps include validation of these devices with lab-based motion capture or pressure-sensitive walkways with patients walking at difference paces to understand any bias based on gait speed introduced due to video sampling rate.^48,49 In home remote gait monitoring is an attractive solution for enhancing and democratizing access to clinical trials and care for pwMS. Evaluating feasibility and validity of in-home monitoring with wearable sensors or depth-camera systems will be required to achieve that goal. While a 20% change in timed 25-foot walk time has been proposed as clinically meaningful,⁵⁰ robust longitudinal studies are needed to determine meaningful differences in other gait metrics.

Our study demonstrates a strong correlation between spatiotemporal measurements obtained from a wearable IMU device and a depth-camera system in pwMS. These metrics are more sensitive to disease-relevant functional changes than the PDDS score, particularly for those with mild disability. These findings highlight the potential for either system to be widely adopted in clinical settings, offering a quantitative and accessible method for evaluating disability and tracking disease progression in MS patients.

Highlights.

• Gait impairment is a measure of disease progression in multiple sclerosis

• Validated gait-assessment tools in multiple sclerosis are critically needed

• Spatiotemporal metrics by depth video correlate with previously validated wearable sensors

• Patient reported disability correlated most strongly with speed

• Study implemented in clinic setting, demonstrating feasibility of practical clinical use