Video-Based Pose Estimation for Gait Analysis in Stroke Survivors during Clinical Assessments: A Proof-of-Concept Study

Recent advancements in deep learning have produced significant progress in markerless human pose estimation, making it possible to estimate human kinematics from single camera videos without the need for reflective markers and specialized labs equipped with motion capture systems. Such algorithms have the potential to enable the quantification of clinical metrics from videos recorded with a handheld camera. Here we used DeepLabCut, an open-source framework for markerless pose estimation, to fine-tune a deep network to track 5 body keypoints (hip, knee, ankle, heel, and toe) in 82 below-waist videos of 8 patients with stroke performing overground walking during clinical assessments. We trained the pose estimation model by labeling the keypoints in 2 frames per video and then trained a convolutional neural network to estimate 5 clinically relevant gait parameters (cadence, double support time, swing time, stance time, and walking speed) from the trajectory of these keypoints. These results were then compared to those obtained from a clinical system for gait analysis (GAITRite®, CIR Systems). Absolute accuracy (mean error) and precision (standard deviation of error) for swing, stance, and double support time were within 0.04 ± 0.11 s; Pearson’s correlation with the reference system was moderate for swing times (<i>r</i> = 0.4–0.66), but stronger for stance and double support time (<i>r</i> = 0.93–0.95). Cadence mean error was −0.25 steps/min ± 3.9 steps/min (<i>r</i> = 0.97), while walking speed mean error was −0.02 ± 0.11 m/s (<i>r</i> = 0.92). These preliminary results suggest that single camera videos and pose estimation models based on deep networks could be used to quantify clinically relevant gait metrics in individuals poststroke, even while using assistive devices in uncontrolled environments. Such development opens the door to applications for gait analysis both inside and outside of clinical settings, without the need of sophisticated equipment.

Longitudinal Changes in Temporospatial Gait Characteristics during the First Year Post-Stroke

Given the paucity of longitudinal data in gait recovery after stroke, we compared temporospatial gait characteristics of stroke patients during subacute (<2 months post-onset, T0) and at approximately 6 and 12 months post-onset (T1 and T2, respectively) and explored the relationship between gait characteristics at T0 and the changes in gait speed from T0 to T1. Forty-six participants were assessed at T0 and a subsample of 24 participants at T2. Outcome measures included Fugl-Meyer lower-extremity motor score, 14 temporospatial gait parameters and symmetry indices of 5 step parameters. Except for step width, all temporospatial parameters improved from T0 to T1 (p ≤ 0.0001). Additionally, significant improvements in symmetry were found for the initial double-support time and single-support time (p ≤ 0.0001). Although group results at T2 were not different from those at T1, the individual analysis revealed that 42% (10/24) of the subsample showed a significant increase in gait speed. The increase in gait speed from T0 to T1 was negatively correlated with gait speed and stride length, and positively correlated with the symmetry indices of stance and single-support times at T0 (p ≤ 0.002). Temporospatial gait parameters and stance time symmetry improve over the first 6 months after stroke with an apparent plateau thereafter. Approximately 40% of the subsample continue to increase gait speed from 6 to 12 months post-stroke. A greater increase in gait speed during the first 6 months post-stroke is associated with initially slower walking, shorter stride length, and more pronounced asymmetry in stance and single-support times. The improvement in lower-extremity motor function and bilateral improvements in step parameters collectively suggest that gait changes over the first 12 months after stroke are likely due to neurological recovery, although some compensation by the non-paretic side cannot be excluded.

THE EFFECTS OF TRAINING ON DIFFERENT SURFACES, ON BALANCE AND GAIT PERFORMANCE IN STROKE HEMIPLEGIA

ABSTRACT Although many studies have focused on balance exercises for elderly or stroke patients, no comprehensive studies have investigated the use of training on different surfaces (TDS) with analysis of gait performance in elderly male stroke patients. The active properties of balance and subjective reporting of functional gait ability were used to identify the effects of TDS. Static balance (SB), dynamic balance (DB) and gait analysis was measured in 30 elderly stroke patients. The patients were divided into the TDS group (n=15) and a control group (CG, n=15). Fifteen elderly stroke patients underwent TDS five times a week for 12 weeks. The data was analyzed using repeated measures analysis of variance. Significant differences were observed between the two groups (TDS and Control): SB (p < 0.0001), DB (OSI: p < 0.0001, APSI: p < 0.001, MLSI: p < 0.004) and gait analysis (right: temporal step time: p < 0.0001, temporal cycle time: p < 0.001, temporal double support time: p < 0.0001; left: temporal step time: p < 0.0001, temporal cycle time: p < 0.0001, temporal double support time: p < 0.0001). TDS in elderly male stroke patients suggests that the characteristics of gait performance in these patients may be improved by increasing static balance, dynamic balance and gait velocity. It is hoped that the results of this trial will provide new information on the effects of TDS on balance stability and gait ability in stroke patients, through changes in stability of the lower extremities. Level III, Case-control Study.

Biomechanism and exercise effect of fitness walking using twin walking sticks

In Japan, walking poles with pairs of sticks developed exclusively for fitness walking have been designed. A new concept of walking style (WS) has been conceived using these walking sticks to “effectively” walk around the city, comprehensive sports parks, or at rehabilitation hospitals. Stick manufacturers are promoting its health benefits; however, evidence supporting these claims is lacking. Hence, this study aimed to measure the influence of walking sticks and evaluate the exercise effect based on functional physical fitness related to WS characteristics. The participants were 12 WS instructors. They engaged in WS at a comfortable speed after walking normally at the same speed (WN) for ∼5 m (seven times), followed by WS again. The walking speed, step length, stride width, walk ratio, one-leg support time, and trajectory of the center of gravity (CG) (in the horizontal and vertical directions of one walking cycle) calculated from the whole-body skeleton model were analyzed. The gait of WS increased the step length, step width, and walking ratio as compared with that of WN (p&lt;0.05). WS likely reduce cadence and one-leg support time (p&lt;0.05). The CG locus in the left-right direction showed no significant differences between WS and WN. The maximum value of the CG locus in the vertical direction was high in WS (p&lt;0.05). WS can be used as a navigation training tool that improves a walker's exercise efficiency and left-right leg coordination, thereby improving walking posture. This may help reduce the anxiety due to injuries and pain that may occur while walking.

Gait and Axial Spondyloarthritis: Comparative Gait Analysis Study Using Foot-Worn Inertial Sensors

Background Axial spondyloarthritis (axSpA) can lead to spinal mobility restrictions associated with restricted lower limb ranges of motion, thoracic kyphosis, spinopelvic ankylosis, or decrease in muscle strength. It is well known that these factors can have consequences on spatiotemporal gait parameters during walking. However, no study has assessed spatiotemporal gait parameters in patients with axSpA. Divergent results have been obtained in the studies assessing spatiotemporal gait parameters in ankylosing spondylitis, a subgroup of axSpA, which could be partly explained by self-reported pain intensity scores at time of assessment. Inertial measurement units (IMUs) are increasingly popular and may facilitate gait assessment in clinical practice. Objective This study compared spatiotemporal gait parameters assessed with foot-worn IMUs in patients with axSpA and matched healthy individuals without and with pain intensity score as a covariate. Methods A total of 30 patients with axSpA and 30 age- and sex-matched healthy controls performed a 10-m walk test at comfortable speed. Various spatiotemporal gait parameters were computed from foot-worn inertial sensors including gait speed in ms–1 (mean walking velocity), cadence in steps/minute (number of steps in a minute), stride length in m (distance between 2 consecutive footprints of the same foot on the ground), swing time in percentage (portion of the cycle during which the foot is in the air), stance time in percentage (portion of the cycle during which part of the foot touches the ground), and double support time in percentage (portion of the cycle where both feet touch the ground). Results Age, height, and weight were not significantly different between groups. Self-reported pain intensity was significantly higher in patients with axSpA than healthy controls (P<.001). Independent sample t tests indicated that patients with axSpA presented lower gait speed (P<.001) and cadence (P=.004), shorter stride length (P<.001) and swing time (P<.001), and longer double support time (P<.001) and stance time (P<.001) than healthy controls. When using pain intensity as a covariate, spatiotemporal gait parameters were still significant with patients with axSpA exhibiting lower gait speed (P<.001), shorter stride length (P=.001) and swing time (P<.001), and longer double support time (P<.001) and stance time (P<.001) than matched healthy controls. Interestingly, there were no longer statistically significant between-group differences observed for the cadence (P=.17). Conclusions Gait was significantly altered in patients with axSpA with reduced speed, cadence, stride length, and swing time and increased double support and stance time. Taken together, these changes in spatiotemporal gait parameters could be interpreted as the adoption of a so-called cautious gait pattern in patients with axSpA. Among factors that may influence gait in patients with axSpA, patient self-reported pain intensity could play a role. Finally, IMUs allowed computation of spatiotemporal gait parameters and are usable to assess gait in patients with axSpA in clinical routine. Trial Registration ClinicalTrials.gov NCT03761212; https://clinicaltrials.gov/ct2/show/NCT03761212 International Registered Report Identifier (IRRID) RR2-10.1007/s00296-019-04396-4

Gait Characteristics and Fatigue Profiles When Standing on Surfaces with Different Hardness: Gait Analysis and Machine Learning Algorithms

Background: Longtime standing may cause fatigue and discomfort in the lower extremities, leading to an increased risk of falls and related musculoskeletal diseases. Therefore, preventive interventions and fatigue detection are crucial. This study aims to explore whether anti-fatigue mats can improve gait parameters following long periods of standing and try to use machine learning algorithms to identify the fatigue states of standing workers objectively. Methods: Eighteen healthy young subjects were recruited to stand on anti-fatigue mats and hard ground to work 4 h, including 10 min rest. The portable gait analyzer collected walking speed, stride length, gait frequency, single support time/double support time, swing work, and leg fall intensity. A Paired sample t-test was used to compare the difference of gait parameters without standing intervention and standing on two different hardness planes for 4 h. An independent sample t-test was used to analyze the difference between males and females. The K-nearest neighbor (KNN) classification algorithm was performed, the subject’s gait characteristics were divided into non-fatigued and fatigue groups. The gait parameters selection and the error rate of fatigue detection were analyzed. Results: When gender differences were not considered, the intensity of leg falling after standing on the hard ground for 4 h was significantly lower than prior to the intervention (p < 0.05). When considering the gender, the stride length and leg falling strength of female subjects standing on the ground for 4 h were significantly lower than those before the intervention (p < 0.05), and the leg falling strength after standing on the mat for 4 h was significantly lower than that recorded before the standing intervention (p < 0.05). The leg falling strength of male subjects standing on the ground for 4 h was significantly lower than before the intervention (p < 0.05). After standing on the ground for 4 h, female subjects’ walking speed and stride length were significantly lower than those of male subjects (p < 0.05). In addition, the accuracy of testing gait parameters to predict fatigue was medium (75%). After standing on the mat was divided into fatigue, the correct rate was 38.9%, and when it was divided into the non-intervention state, the correct rate was 44.4%. Conclusion: The results show that the discomfort and fatigue caused by standing for 4 h could lead to the gait parameters variation, especially in females. The use of anti-fatigue mats may improve the negative influence caused by standing for a long period. The results of the KNN classification algorithm showed that gait parameters could be identified after fatigue, and the use of an anti-fatigue mat could improve the negative effect of standing for a long time. The accuracy of the prediction results in this study was moderate. For future studies, researchers need to optimize the algorithm and include more factors to improve the prediction accuracy.

Season- and Trend-aware Symbolic Approximation for Accurate and Efficient Time Series Matching

Processing and analyzing time series datasets have become a central issue in many domains requiring data management systems to support time series as a native data type. A core access primitive of time series is matching, which requires efficient algorithms on-top of appropriate representations like the symbolic aggregate approximation (SAX) representing the current state of the art. This technique reduces a time series to a low-dimensional space by segmenting it and discretizing each segment into a small symbolic alphabet. Unfortunately, SAX ignores the deterministic behavior of time series such as cyclical repeating patterns or a trend component affecting all segments, which may lead to a sub-optimal representation accuracy. We therefore introduce a novel season- and a trend-aware symbolic approximation and demonstrate an improved representation accuracy without increasing the memory footprint. Most importantly, our techniques also enable a more efficient time series matching by providing a match up to three orders of magnitude faster than SAX.

Finite Analysis of Support Time for Tunnels in Strain-softening Rock Mass in Consideration of Fictitious Support Pressure

This study considers that the support time is quantitatively determined by the production limit of the displacement reduction factor and the support force under the extrusion conditions of the strain-softening rock mass. Therefore, the two indicators of the downlines under the support time are the displacement reduction factor of the support force and the yield limit. Based on the solution of the fictitious pressure proposed in an existing paper, the finite difference method is adopted to investigate the variations of the support force and displacement reduction factor versus the delayed distance considering different support types, initial stresses, and post-peak behaviours. The results show that on the one hand, the delay distance is suggested within 1 R0 in most tunnel cases; on the other hand, the factors have greater impact on rock-support interactions are rock mass and in-situ stress. Relatively contrast, softening and expansion behavior was not significant enough. Furthermore, it is also very important in composite support systems to assess the proportion of loads shared with the weakest part.

Assessment of gait in patients with fibromyalgia during motor and cognitive dual task walking: a cross-sectional study

Background The aim of the study was to assess gait pattern of patients diagnosed with fibromyalgia (FM) while performing demanding motor and/or cognitive dual tasks while walking. Further, idea was to explore possible correlations of dual task gait pattern alterations to patients’ functional status and presence or absence of clinical symptoms associated with FM. Methods Twenty-four female FM patients and 24 healthy female subjects performed a basic walking task, a dual motor, a dual mental (cognitive) and a combined, dual motor and cognitive task simultaneously. Quantitative spatial (stride length) and temporal (cycle time, swing time and double support time) gait parameters were measured using GAITRite walkway system and their variability was assessed. Patients underwent clinical examination including assessment of functional status, pain and fatigue level, psychiatric and cognitive manifestations. Results The motor, cognitive and combined dual tasks affect gait performance in FM patients. Difference in tasks between FM and healthy subjects was found as double support time prolongation. Comparison of tasks showing that cycle time in FM was longer than controls and stride length was shorter in patients for all conditions, while no changes were found in any of the gait parameters variability. Further, mental/cognitive dual tasks had a larger effect than motor tasks. Correlations were also found between depression and functional status of the patients and the gait parameters. Conclusions Gait is affected in FM patients while dual task walking. No changes in stride-to-stride variability point that patients preserve stability in complex walking situations. Analysis of gait may provide additional information for the FM identification based on presence of clinical features and cognitive status. Correlation of dual task gait alterations with occurrence of clinical symptoms and influence of cognitive changes on gait pattern could additionally define FM subgroups.