Exploration and demonstration of explainable machine learning models in prosthetic rehabilitation-based gait analysis.

Quantitative gait analysis is important for understanding the non-typical walking patterns associated with mobility impairments. Conventional linear statistical methods and machine learning (ML) models are commonly used to assess gait performance and related changes in the gait parameters. Nonetheless, explainable machine learning provides an alternative technique for distinguishing the significant and influential gait changes stemming from a given intervention. The goal of this work was to demonstrate the use of explainable ML models in gait analysis for prosthetic rehabilitation in both population- and sample-based interpretability analyses. Models were developed to classify amputee gait with two types of prosthetic knee joints. Sagittal plane gait patterns of 21 individuals with unilateral transfemoral amputations were video-recorded and 19 spatiotemporal and kinematic gait parameters were extracted and included in the models. Four ML models-logistic regression, support vector machine, random forest, and LightGBM-were assessed and tested for accuracy and precision. The Shapley Additive exPlanations (SHAP) framework was applied to examine global and local interpretability. Random Forest yielded the highest classification accuracy (98.3%). The SHAP framework quantified the level of influence of each gait parameter in the models where knee flexion-related parameters were found the most influential factors in yielding the outcomes of the models. The sample-based explainable ML provided additional insights over the population-based analyses, including an understanding of the effect of the knee type on the walking style of a specific sample, and whether or not it agreed with global interpretations. It was concluded that explainable ML models can be powerful tools for the assessment of gait-related clinical interventions, revealing important parameters that may be overlooked using conventional statistical methods.

Three-Dimensional Instrumented Gait Analysis for Children With Cerebral Palsy: An Evidence-Based Clinical Practice Guideline.

BACKGROUND: Children with cerebral palsy (CP) who walk have complex gait patterns and deviations often requiring physical therapy (PT)/medical/surgical interventions. Walking in children with CP can be assessed with 3-dimensional instrumented gait analysis (3D-IGA) providing kinematics (joint angles), kinetics (joint moments/powers), and muscle activity. PURPOSE: This clinical practice guideline provides PTs, physicians, and associated clinicians involved in the care of children with CP, with 7 action statements on when and how 3D-IGA can inform clinical assessments and potential interventions. It links the action statement grades with specific levels of evidence based on a critical appraisal of the literature. CONCLUSIONS: This clinical practice guideline addresses 3D-IGA's utility to inform surgical and non-surgical interventions, to identify gait deviations among segments/joints and planes and to evaluate the effectiveness of interventions. Best practice statements provide guidance for clinicians about the preferred characteristics of 3D-IGA laboratories including instrumentation, staffing, and reporting practices.Video Abstract: Supplemental digital content available at http://links.lww.com/PPT/A524.

Demonstrating the utility of Instrumented Gait Analysis in the treatment of children with cerebral palsy.

BACKGROUND: Instrumented gait analysis (IGA) has been around for a long time but has never been shown to be useful for improving patient outcomes. In this study we demonstrate the potential utility of IGA by showing that machine learning models are better able to estimate treatment outcomes when they include both IGA and clinical (CLI) features compared to when they include CLI features alone. DESIGN: We carried out a retrospective analysis of data from ambulatory children diagnosed with cerebral palsy who were seen at least twice at our gait analysis center. Individuals underwent a variety of treatments (including no treatment) between sequential gait analyses. We fit Bayesian Additive Regression Tree (BART) models that estimated outcomes for mean stance foot progression to demonstrate the approach. We built two models: one using CLI features only, and one using CLI and IGA features. We then compared the models' performance in detail. We performed similar, but less detailed, analyses for a number of other outcomes. All results were based on independent test data from a 70%/30% training/testing split. RESULTS: The IGA model was more accurate than the CLI model for mean stance-phase foot progression outcomes (RMSEIGA = 11∘, RMSECLI = 13∘) and explained more than 1.5 × as much of the variance (R2IGA = .45, R2CLI = .28). The IGA model outperformed the CLI model for every level of treatment complexity, as measured by number of simultaneous surgeries. The IGA model also exhibited superior performance for estimating outcomes of mean stance-phase knee flexion, mean stance-phase ankle dorsiflexion, maximum swing-phase knee flexion, gait deviation index (GDI), and dimensionless speed. INTERPRETATION: The results show that IGA has the potential to be useful in the treatment planning process for ambulatory children diagnosed with cerebral palsy. We propose that the results of machine learning outcome estimators-including estimates of uncertainty-become the primary IGA tool utilized in the clinical process, complementing the standard medical practice of conducting a through patient history and physical exam, eliciting patient goals, reviewing relevant imaging data, and so on.

Gait differences between COPD and healthy controls: systematic review and meta-analysis.

BACKGROUND: Despite the importance of gait as a determinant of falls, disability and mortality in older people, understanding of gait impairment in COPD is limited. This study aimed to identify differences in gait characteristics during supervised walking tests between people with COPD and healthy controls. METHODS: We searched 11 electronic databases, supplemented by Google Scholar searches and manual collation of references, in November 2019 and updated the search in July 2021. Record screening and information extraction were performed independently by one reviewer and checked for accuracy by a second. Meta-analyses were performed in studies not considered at a high risk of bias. RESULTS: Searches yielded 21 085 unique records, of which 25 were included in the systematic review (including 1015 people with COPD and 2229 healthy controls). Gait speed was assessed in 17 studies (usual speed: 12; fast speed: three; both speeds: two), step length in nine, step duration in seven, cadence in six, and step width in five. Five studies were considered at a high risk of bias. Low-quality evidence indicated that people with COPD walk more slowly than healthy controls at their usual speed (mean difference (MD) -19 cm·s-1, 95% CI -28 to -11 cm·s-1) and at a fast speed (MD -30 cm·s-1, 95% CI -47 to -13 cm·s-1). Alterations in other gait characteristics were not statistically significant. CONCLUSION: Low-quality evidence shows that people with COPD walk more slowly than healthy controls, which could contribute to an increased falls risk. The evidence for alterations in spatial and temporal components of gait was inconclusive. Gait impairment appears to be an important but understudied area in COPD.

Identification and interpretation of gait analysis features and foot conditions by explainable AI.

Clinical gait analysis is a crucial step for identifying foot disorders and planning surgery. Automating this process is essential for efficiently assessing the substantial amount of gait data. In this study, we explored the potential of state-of-the-art machine learning (ML) and explainable artificial intelligence (XAI) algorithms to automate all various steps involved in gait analysis for six specific foot conditions. To address the complexity of gait data, we manually created new features, followed by recursive feature elimination using Support Vector Machines (SVM) and Random Forests (RF) to eliminate low-variance features. SVM, RF, K-nearest Neighbor (KNN), and Logistic Regression (LREGR) were compared for classification, with a Majority Voting (MV) model combining trained models. KNN and MV achieved mean balanced accuracy, recall, precision, and F1 score of 0.87. All models were interpreted using Local Interpretable Model-agnostic Explanation (LIME) method and the five most relevant features were identified for each foot condition. High success scores indicate a strong relationship between selected features and foot conditions, potentially indicating clinical relevance. The proposed ML pipeline, adaptable for other foot conditions, showcases its potential in aiding experts in foot condition identification and planning surgeries.

Three-Dimensional Gait Analysis as a Biomarker for GTP Cyclohydrolase 1-Deficient Dopa-Responsive Dystonia.

BACKGROUND: GTP-cyclohydrolase 1-deficient dopa-responsive dystonia (GTPCH1-deficient DRD) typically presents in childhood with dystonic posture of the lower extremities, gait impairment, and a significant response to levodopa. We performed three-dimensional gait analysis (3DGA) to quantitatively assess the gait characteristics and changes associated with levodopa treatment in patients with GTPCH1-deficient DRD. METHODS: Three levodopa-treated patients with GTPCH1-deficient DRD underwent 3DGA twice, longitudinally. Changes were evaluated for cadence; gait speed; step length; gait deviation index; kinematic data of the pelvis, hip, knee, and ankle joints; and foot progression angle. RESULTS: Levodopa treatment increased the cadence and gait speed in one of three patients and increased the gait deviation index in two of three patients. The kinematic data for each joint exhibited different characteristics, with some improvement observed in each of the three patients. There was consistent marked improvement in the abnormal foot progression angle; one patient had excessive external rotation of one foot, another had excessive bilateral internal rotation, and the other had excessive internal rotation of one foot and excessive external rotation of the opposite foot, all of which improved. CONCLUSION: The 3DGA findings demonstrate that the gait pathology and recovery process in GTPCH1-deficient DRD vary from case to case. Changes in the foot progression angle and gait deviation index can enable the effects of treatment to be more easily evaluated.

Biomechanical gait analysis and rehabilitation in a traumatic hallux deficit patient: a case report.

BACKGROUND: The hallux plays a crucial role in maintaining standing balance and facilitating forward and backward movements during gait. CASE PRESENTATION: A 21-year-old Japanese patient, suffering from a traumatic hallux deficit with only a portion of the basal phalanx intact, underwent rehabilitation treatment. The thenar area exhibited instability, leading to impaired balance and walking difficulties. Biomechanical assessment revealed the need for a rehabilitation strategy for the foot, as well as the knee, hip, and trunk. A rehabilitation protocol was designed to enhance medial foot loading during walking and standing, including balance and trunk strength training. After a 12-week rehabilitation period, the patient's gait showed significant improvement. Specifically, the load response and single-support phases of the gait cycle on the affected side increased from 46.9% to 49.3%, while the pre-swing phase decreased from 14.6% to 11.6%. The vertical component of the ground reaction force rose from 599.8 to 647.5 N. The enhanced stability from balance training and increased muscle strength contributed to the patient's improved walking and balance. CONCLUSION: A patient with a traumatic hallux deficit underwent conservative treatment through strategic rehabilitation according to biomechanical assessment. This case report underscores the value of biomechanical gait analysis in the conservative management of similar conditions.

Improving the reliability of underwater gait analysis using wearable pressure and inertial sensors.

This work addresses the lack of reliable wearable methods to assess walking gaits in underwater environments by evaluating the lateral hydrodynamic pressure exerted on lower limbs. Sixteen healthy adults were outfitted with waterproof wearable inertial and pressure sensors. Gait analysis was conducted on land in a motion analysis laboratory using an optoelectronic system as reference, and subsequently underwater in a rehabilitation swimming pool. Differences between the normalized land and underwater gaits were evaluated using temporal gait parameters, knee joint angles and the total water pressure on the lower limbs. The proposed method was validated against the optoelectronic system on land; gait events were identified with low bias (0.01s) using Bland-Altman plots for the stride time, and an acceptable error was observed when estimating the knee angle (10.96° RMSE, Bland-Altman bias -2.94°). The kinematic differences between the land and underwater environments were quantified, where it was observed that the temporal parameters increased by more than a factor of two underwater (p<0.001). The subdivision of swing and stance phases remained consistent between land and water trials. A higher variability of the knee angle was observed in water (CV = 60.75%) as compared to land (CV = 31.02%). The intra-subject variability of the hydrodynamic pressure on the foot ([Formula: see text] = 39.65%) was found to be substantially lower than that of the knee angle (CVz = 67.69%). The major finding of this work is that the hydrodynamic pressure on the lower limbs may offer a new and more reliable parameter for underwater motion analysis as it provided a reduced intra-subject variability as compared to conventional gait parameters applied in land-based studies.

Inter-trial variability is higher in 3D markerless compared to marker-based motion capture: Implications for data post-processing and analysis.

Markerless motion capture has recently attracted significant interest in clinical gait analysis and human movement science. Its ease of use and potential to streamline motion capture recordings bear great potential for out-of-the-laboratory measurements in large cohorts. While previous studies have shown that markerless systems can achieve acceptable accuracy and reliability for kinematic parameters of gait, they also noted higher inter-trial variability of markerless data. Since increased inter-trial variability can have important implications for data post-processing and analysis, this study compared the inter-trial variability of simultaneously recorded markerless and marker-based data. For this purpose, the data of 18 healthy volunteers were used who were instructed to simulate four different gait patterns: physiological, crouch, circumduction, and equinus gait. Gait analysis was performed using the smartphone-based markerless system OpenCap and a marker-based motion capture system. We compared the inter-trial variability of both systems and also evaluated if changes in inter-trial variability may depend on the analyzed gait pattern. Compared to the marker-based data, we observed an increase of inter-trial variability for the markerless system ranging from 6.6% to 22.0% for the different gait patterns. Our findings demonstrate that the markerless pose estimation pipelines can introduce additionally variability in the kinematic data across different gait patterns and levels of natural variability. We recommend using averaged waveforms rather than single ones to mitigate this problem. Further, caution is advised when using variability-based metrics in gait and human movement analysis based on markerless data as increased inter-trial variability can lead to misleading results.

Markerless gait analysis through a single camera and computer vision.

The assessment of gait performance using quantitative measures can yield crucial insights into an individual's health status. Recently, computer vision-based human pose estimation has emerged as a promising solution for markerless gait analysis, as it allows for the direct extraction of gait parameters from videos. This study aimed to compare the lower extremity kinematics and spatiotemporal gait parameters obtained from a single-camera-based markerless method with those acquired from a marker-based motion tracking system across a healthy population. Additionally, we investigated the impact of camera viewing angles and distances on the accuracy of the markerless method. Our findings demonstrated a robust correlation and agreement (Rxy > 0.75, Rc > 0.7) between the markerless and marker-based methods for most spatiotemporal gait parameters. We also observed strong correlations (Rxy > 0.8) between the two methods for hip flexion/extension, knee flexion/extension, hip abduction/adduction, and hip internal/external rotation. Statistical tests revealed significant effects of viewing angles and distances on the accuracy of the identified gait parameters. While the markerless method offers an alternative for general gait analysis, particularly when marker use is impractical, its accuracy for clinical applications remains insufficient and requires substantial improvement. Future investigations should explore the potential of the markerless system to measure gait parameters in pathological gaits.

Regional differences in the biological variability of plantar pressure as a basis for refining diagnostic gait analysis.

The variability of movement plays a crucial role in shaping individual's gait pattern and could, therefore, potentially serve diagnostic purposes. Nevertheless, existing concepts for the use of variability in diagnosing gait present a challenge due to the lack of adequate benchmarks and methods for comparison. We assessed the individuality of contribution of foot parts that directly mediate the transmission of forces between the foot and the ground in body weight shifting during walking based on 200 pedobarometric measurements corresponding to the analysed foot parts for each of 19 individuals in a homogeneous study group. Our results show a degree of individualisation of the contribution of particular foot parts in the weight-shift high enough to justify the need to consider it in the diagnostic analysis. Furthermore they reveal noticeable, functionally driven differences between plantar areas most apparent between the lowest individuality for the first foot ray and the highest for second one and metatarsus. The diagnostic reference standard in pedobarometry should describe the contribution in the shift of body weight during walking for each area of the foot separately and include information on the intra-individual variation and individualisation of descriptors of the contribution. Such a comprehensive standard has the potential to increase the diagnostic value of pedobarometry through enrichment of the assessment description.

Evaluation of variability in performance and paw placement patterns by dogs completing the dog walk obstacle in an agility competition.

The objective of this study was to describe paw placement patterns for canine athletes completing the dog walk obstacle during canine agility trials. It was hypothesized that dogs would demonstrate defined sets of paw placement patterns as they complete the dog walk obstacle and that those could be classified based on end contact behavior. Videos of 296 dogs attempting the dog walk obstacle at the 2021 UK Agility International (UKI) US Open were reviewed online. Data observed from video evaluation included front and rear limb paw placement across the dog walk and time to complete the obstacle. Results showed a high variability in obstacle performance. Mean time to complete the entire obstacle was 2.26 seconds (sd = 1.03). Mean and median completion times were qualitatively similar across all height classes. A slight majority of dogs hit the up ramp with their right foot first indicating running on their left lead (n = 185, 63%) with some variation observed between heights. Likewise, a slight majority (58%) of dogs hit the down ramp with their right front foot first (151/262). Given the high variation in completion times and paw placements, we could not identify clear patterns of dog walk performance. The large amount of variation observed with the dog walk obstacle suggests a need for future studies to employ alternative methods for objective gait analysis and to strategically select dogs to reflect the large variety in obstacle performance observed here.

Executive Functions Assessment Based on Wireless EEG and 3D Gait Analysis During Dual-Task: A Feasibility Study.

Executive functions (EFs) are neurocognitive processes planning and regulating daily life actions. Performance of two simultaneous tasks, requiring the same cognitive resources, lead to a cognitive fatigue. Several studies investigated cognitive-motor task and the interference during walking, highlighting an increasing risk of falls especially in elderly and people with neurological diseases. A few studies instrumentally explored relationship between activation-no-activation of two EFs (working memory and inhibition) and spatial-temporal gait parameters. Aim of our study was to detect activation of inhibition and working memory during progressive difficulty levels of cognitive tasks and spontaneous walking using, respectively, wireless electroencephalography (EEG) and 3D-gait analysis. Thirteen healthy subjects were recruited. Two cognitive tasks were performed, activating inhibition (Go-NoGo) and working memory (N-back). EEG features (absolute and relative power in different bands) and kinematic parameters (7 spatial-temporal ones and Gait Variable Score for 9 range of motion of lower limbs) were analyzed. A significant decrease of stride length and an increase of external-rotation of foot progression were found during dual task with Go-NoGo. Moreover, a significant correlation was found between the relative power in the delta band at channels Fz, C4 and progressive difficulty levels of Go-NoGo (activating inhibition) during walking, whereas working memory showed no correlation. This study reinforces the hypothesis of the prevalent involvement of inhibition with respect to working memory during dual task walking and reveals specific kinematic adaptations. The foundations for EEG-based monitoring of cognitive processes involved in gait are laid. Clinical and Translational Impact Statement: Clinical and instrumental evaluation and training of executive functions (as inhibition), during cognitive-motor task, could be useful for rehabilitation treatment of gait disorder in elderly and people with neurological disease.

Gait disorder classification based on effective feature selection and unsupervised methodology.

In gait stability analysis, patients suffering from dysfunction problems are impacted by shifts in their dynamic balance. Monitoring the patients' progress is important for allowing physicians and patients to observe the rehabilitation process accurately. In this study, we designed a new methodology for classifying gait disorders to quantify patients' progress. The dataset in this study includes 84 measurements of 37 patients based on a physician's opinion. In this study, the system, which includes a Kinect camera to observe and store the frames of patients walking down a hallway, a key-point detector to detect the skeletal key points, and an encoder transformer classifier network integrated with generator-discriminator networks (ET-GD), is designed to evaluate the classification of gait dysfunction. The detector extracts the skeletal key points of patients. After feature engineering, the selected high-level features are fed into the proposed neural network to analyse patient movement and perform the final evaluation of gait dysfunction. The proposed network is inspired by the 1D encoder transformer, which is integrated with two main networks: a network for classification and a network to generate fake output data similar to the input data. Furthermore, we used a discriminator structure to distinguish between the actual data (input) and fake data (generated data). Due to the multi-structural networks in the proposed method, multi-loss functions need to be optimised; this increases the accuracy of the encoder transformer classifier.

Gait analysis of a patient after femoral nerve and malignant soft tissue tumor resections: a case report.

BACKGROUND: Malignant femoral soft tissue tumors are occasionally resected together with the femoral nerves, but this can cause loss of knee extensor muscle activity. To the best of our knowledge, no previous reports have detailed the gait analysis of such cases in combination with electromyography. Herein, we report the gait analysis of a patient who underwent left groin synovial sarcoma and left femoral nerve resection 12 years ago. CASE PRESENTATION: We analyzed the gait of a 38-year-old man who was able to walk unaided after the resection of a synovial sarcoma in the left groin together with the ipsilateral femoral nerve. The muscle activities of the affected medial (MH) and lateral hamstrings (LH), and lateral heads of the gastrocnemius (GL) were increased during 50-75% of the stance phase. The hip flexion angle of the affected limb was smaller, and the ankle plantar flexion angle of the affected limb was larger than that of the non-affected limb. This means that in the affected limb, the hip and ankle angles were adjusted to prevent knee collapse, and the MH, LH, and GL muscles contributed in the mid- and late-stance phases. Moreover, we found that the hamstring and gastrocnemius of the affected limb worked together to keep the ipsilateral knee extended in the mid-stance phase and slightly flexed in the late-stance phase. CONCLUSIONS: Patients capable of walking after femoral nerve resection may control their hamstrings and gastrocnemius muscles collaboratively to prevent ipsilateral knee collapse in the mid- and late-stance phases.

Estimation of Gait Parameters for Adults with Surface Electromyogram Based on Machine Learning Models.

Gait analysis has been studied over the last few decades as the best way to objectively assess the technical outcome of a procedure designed to improve gait. The treating physician can understand the type of gait problem, gain insight into the etiology, and find the best treatment with gait analysis. The gait parameters are the kinematics, including the temporal and spatial parameters, and lack the activity information of skeletal muscles. Thus, the gait analysis measures not only the three-dimensional temporal and spatial graphs of kinematics but also the surface electromyograms (sEMGs) of the lower limbs. Now, the shoe-worn GaitUp Physilog® wearable inertial sensors can easily measure the gait parameters when subjects are walking on the general ground. However, it cannot measure muscle activity. The aim of this study is to measure the gait parameters using the sEMGs of the lower limbs. A self-made wireless device was used to measure the sEMGs from the vastus lateralis and gastrocnemius muscles of the left and right feet. Twenty young female subjects with a skeletal muscle index (SMI) below 5.7 kg/m2 were recruited for this study and examined by the InBody 270 instrument. Four parameters of sEMG were used to estimate 23 gait parameters. They were measured using the GaitUp Physilog® wearable inertial sensors with three machine learning models, including random forest (RF), decision tree (DT), and XGBoost. The results show that 14 gait parameters could be well-estimated, and their correlation coefficients are above 0.800. This study signifies a step towards a more comprehensive analysis of gait with only sEMGs.

A Transferable Lidar-Based Method to Conduct Contactless Assessments of Gait Parameters in Diverse Home-like Environments.

Gait abnormalities in older adults are linked to increased risks of falls, institutionalization, and mortality, necessitating accurate and frequent gait assessments beyond traditional clinical settings. Current methods, such as pressure-sensitive walkways, often lack the continuous natural environment monitoring needed to understand an individual's gait fully during their daily activities. To address this gap, we present a Lidar-based method capable of unobtrusively and continuously tracking human leg movements in diverse home-like environments, aiming to match the accuracy of a clinical reference measurement system. We developed a calibration-free step extraction algorithm based on mathematical morphology to realize Lidar-based gait analysis. Clinical gait parameters of 45 healthy individuals were measured using Lidar and reference systems (a pressure-sensitive walkway and a video recording system). Each participant participated in three predefined ambulation experiments by walking over the walkway. We observed linear relationships with strong positive correlations (R2>0.9) between the values of the gait parameters (step and stride length, step and stride time, cadence, and velocity) measured with the Lidar sensors and the pressure-sensitive walkway reference system. Moreover, the lower and upper 95% confidence intervals of all gait parameters were tight. The proposed algorithm can accurately derive gait parameters from Lidar data captured in home-like environments, with a performance not significantly less accurate than clinical reference systems.

Special Considerations in Podiatric Science: Translational Research, Cadavers, Gait Analysis, Dermatology, and Databases.

The objective of this article is to provide a brief overview of the critical analysis and design of unique and perhaps less common methodologies in podiatric science. These include basic science translational designs, cadaveric investigations, gait analyses, dermatologic studies, and database analysis. The relative advantages, disadvantages, and inherent limitations are reviewed with an intention to improve the interpretation of results and advance future foot and ankle scientific endeavors.

Validation of portable in-clinic video-based gait analysis for prosthesis users.

Despite the common focus of gait in rehabilitation, there are few tools that allow quantitatively characterizing gait in the clinic. We recently described an algorithm, trained on a large dataset from our clinical gait analysis laboratory, which produces accurate cycle-by-cycle estimates of spatiotemporal gait parameters including step timing and walking velocity. Here, we demonstrate this system generalizes well to clinical care with a validation study on prosthetic users seen in therapy and outpatient clinics. Specifically, estimated walking velocity was similar to annotated 10-m walking velocities, and cadence and foot contact times closely mirrored our wearable sensor measurements. Additionally, we found that a 2D keypoint detector pretrained on largely able-bodied individuals struggles to localize prosthetic joints, particularly for those individuals with more proximal or bilateral amputations, but after training a prosthetic-specific joint detector video-based gait analysis also works on these individuals. Further work is required to validate the other outputs from our algorithm including sagittal plane joint angles and step length. Code for the gait transformer and the trained weights are available at https://github.com/peabody124/GaitTransformer .

Systematic review of automatic post-stroke gait classification systems.

BACKGROUND: Gait classification is a clinically helpful task performed after a stroke in order to guide rehabilitation therapy. Gait disorders are commonly identified using observational gait analysis in clinical settings, but this approach is limited due to low reliability and accuracy. Data-driven gait classification can quantify gait deviations and categorise gait patterns automatically possibly improving reliability and accuracy; however, the development and clinical utility of current data driven systems has not been reviewed previously. RESEARCH QUESTION: The purpose of this systematic review is to evaluate the literature surrounding the methodology used to develop automatic gait classification systems, and their potential effectiveness in the clinical management of stroke-affected gait. METHOD: The database search included PubMed, IEEE Xplore, and Scopus. Twenty-one studies were identified through inclusion and exclusion criteria from 407 available studies published between 2015 and 2022. Development methodology, classification performance, and clinical utility information were extracted for review. RESULTS AND SIGNIFICANCE: Most of gait classification systems reported a classification accuracy between 80%-100%. However, collated studies presented methodological errors in machine learning (ML) model development. Further, many studies neglected model components such as clinical utility (e.g., predictions don't assist clinicians or therapists in making decisions, interpretability, and generalisability). We provided recommendations to guide development of future post-stroke automatic gait classification systems to better assist clinicians and therapists. Future automatic gait classification systems should emphasise the clinical significance and adopt a standardised development methodology of ML model.