Foot Off Research Articles

Robust deep learning-based gait event detection across various pathologies.

The correct estimation of gait events is essential for the interpretation and calculation of 3D gait analysis (3DGA) data. Depending on the severity of the underlying pathology and the availability of force plates, gait events can be set either manually by trained clinicians or detected by automated event detection algorithms. The downside of manually estimated events is the tedious and time-intensive work which leads to subjective assessments. For automated event detection algorithms, the drawback is, that there is no standardized method available. Algorithms show varying robustness and accuracy on different pathologies and are often dependent on setup or pathology-specific thresholds. In this paper, we aim at closing this gap by introducing a novel deep learning-based gait event detection algorithm called IntellEvent, which shows to be accurate and robust across multiple pathologies. For this study, we utilized a retrospective clinical 3DGA dataset of 1211 patients with four different pathologies (malrotation deformities of the lower limbs, club foot, infantile cerebral palsy (ICP), and ICP with only drop foot characteristics) and 61 healthy controls. We propose a recurrent neural network architecture based on long-short term memory (LSTM) and trained it with 3D position and velocity information to predict initial contact (IC) and foot off (FO) events. We compared IntellEvent to a state-of-the-art heuristic approach and a machine learning method called DeepEvent. IntellEvent outperforms both methods and detects IC events on average within 5.4 ms and FO events within 11.3 ms with a detection rate of ≥ 99% and ≥ 95%, respectively. Our investigation on generalizability across laboratories suggests that models trained on data from a different laboratory need to be applied with care due to setup variations or differences in capturing frequencies.

Feature Selection using Random Forest Classifier for Foot Strike Event Detection in Toe Walkers

Automated Gait event identification of Foot Strike (FS) and Foot Off (FO) in pathological gait data, can be time saving in comparison to conventional manual annotations done currently. Identification of FS and FO allows breaking walking trials into gait cycles and hence aids in comparison of gait parameters like joint angles, forces and moments across gait cycles. Automated Gait Event Detection is also useful in development of wearable sensor devices and robotic systems that assist gait. Researchers have proposed several automatic gait event detection algorithms based on kinematic parameters and systematic study of the literature suggests specific parameters to have higher contribution in identification of FS event in all common pathological gait patterns. We used Random Forest Classifier Feature selection technique to identify high contributing features in FS event in toe walking pediatric pathological gait dataset and the results suggest high similarity in selected features by the machine learning technique with those suggested by popular event detection algorithms based on kinematic parameters for pathological gait. Hence we conclude that RFC feature selection is suitable for feature selection in toe walkers gait dataset for event detection purpose. Keyword : Feature selection, foot off, foot strike, pathological gait.

Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis-Proof of Concept.

Gait phase detection in IMU-based gait analysis has some limitations due to walking style variations and physical impairments of individuals. Therefore, available algorithms may not work properly when the gait data is noisy, or the person rarely reaches a steady state of walking. The aim of this work was to employ Artificial Intelligence (AI), specifically a long short-term memory (LSTM) algorithm, to overcome these weaknesses. Three supervised LSTM-based models were designed to estimate the expected gait phases, including foot-off (FO), mid-swing (MidS) and foot-contact (FC). For collecting gait data two tri-axial inertial sensors were located above each ankle. The angular velocity magnitude, rotation matrix magnitude and free acceleration magnitude were captured for data labeling and turning detection and to strengthen the model, respectively. To do so, a train dataset based on a novel movement protocol was acquired. A validation dataset similar to a train dataset was generated as well. Five test datasets from already existing data were also created to independently evaluate the models. After testing the models on validation and test datasets, all three models demonstrated promising performance in estimating desired gait phases. The proposed approach proves the possibility of employing AI-based algorithms to predict labeled gait phases from a time series of gait data.

A new deep learning-based method for the detection of gait events in children with gait disorders: Proof-of-concept and concurrent validity

The stance and swing phases of the gait cycle are defined by foot strike (FS) and foot off (FO). Accurate determination of these events is thus an essential component of 3D motion recordings processing. Several methods have been developed for the automatic detection of these events (based on the heuristics of 3D marker position, velocity and acceleration), however the results may be inaccurate due to the high variability that is intrinsic to pathological gait. For this reason, gait events are still commonly determined manually, which is a tedious process. Here we propose a new application (DeepEvent) of a long short term memory recurrent neural network for the automatic detection of gait events. The 3D position and velocity of the markers on the heel, toe and lateral malleolus were used by the network to determine FS and FO. The method was developed from 10526 FS and 9375 FO from 226 children. DeepEvent predicted FS within 5.5 ms and FO within 10.7 ms of the gold standard (automatic determination using force platform data) and was more accurate than common heuristic marker trajectory-based methods proposed in the literature and another deep learning method.A sensitivity analysis showed that DeepEvent mainly used the toe and heel markers (z-axis (longitudinal) position and velocity) at the beginning and end of gait cycle to predict FS, and the toe marker (x-axis (anterior/posterior) velocity and z-axis position and velocity) at around 60% of the gait cycle to predict FO.

The validity of spatiotemporal gait analysis using dual laser range sensors: a cross-sectional study

BackgroundThe spatiotemporal parameters were used for sophisticated gait analysis in widespread clinical use. Recently, a laser range sensor has been proposed as a new device for the spatiotemporal gait measurement. However, measurement using a single laser range sensor can only be used for short-range gait measurements because the device irradiates participants with lasers in a radial manner. For long-range gait measurement, the present study uses a modified method using dual laser range sensors installed at opposite ends of the walking path. The aim of present study was to investigate the concurrent validity of the proposed method for spatiotemporal gait measurement by comparison to a computer-based instrumented walkway system.MethodsTen healthy participants were enrolled in this study. Ten-meter walking tests at 100, 75, and 50% of the comfortable speed were conducted to determine the concurrent validity of the proposed method compared to instrumented walkway measurements. Frequency distributions of errors for foot-contact (FC) and foot-off (FO) estimated times between the two systems were also calculated to determine the adequacy of estimation of FC and FO from three perspectives: accuracy (smallness of mean error), precision (smallness of variability), and unambiguity (monomodality of histogram). Intra-class correlation coefficient (2,1) was used to determine the concurrent validity of spatiotemporal parameters between the two systems.ResultThe results indicate that the detection times for FC and FO estimated by the proposed method did not differ from those measured by the instrumented walkway reference system. In addition, histogram for FC and FO showed monomodality. Intra-class correlation coefficients of the spatiotemporal parameters (stance time: 0.74; double support time: 0.56; stride time: 0.89; stride length: 0.83; step length: 0.71; swing time: 0.23) were not high enough. The mean errors of all spatiotemporal parameters were small.ConclusionsThese results suggest that the proposed lacks sufficient concurrent validity for spatiotemporal gait measurement. Further improvement of this proposed system seems necessary.Trial registrationUMIN000032710. Registered 24 May 2018. Retrospectively registered.

Development of an embedded gait analysis system for brain damaged patients

Optical motion analysis systems are widely used to evaluate the efficiency of treatment/rehabilitation for poststroke patients. These systems are only available in clinical settings due to their size/cost and need of complicate post processing. This paper presents the development of an embedded low cost and easy-to-use system for gait analysis, which measures the joints angles and automatically detects foot-off (FO) and foot-contact (FC) events of stroke walking. The system uses 7 wireless inertial measurement unit (IMUs) to measure the body segments orientations and 2 ultra-wideband (UWB) rangefinders placed near the heels for foot relative distance measurement. Data from sensors are synchronized and recorded in SD card at 70 Hz by controller (AtoutNovation, France). The locomotion of 14 stroke and 4 healthy subjects have been evaluated simultaneously with proposed system and a 3D motion analysis system. Eight gait records at a self-selected speed along 8 m walkway were performed by each subject. The joint angles in sagittal plane were calculated using orientations of segments. To avoid perturbation of external magnetic field (e.g. treadmill walking), the magnetometer was not used, therefore the system cannot calculate 3D joints values. Relative feet distance and foot angular velocity were used to determinate gait events. The joints angles in sagittal plane are segmented, normalized and averaged in gait cycle automatically based on detected gait events. The comparisons between proposed system and motion capture system showed a precision of 28.6 ± 28.3 ms for FC detection and 18.1 ± 18.0 for FO events. 98.9% of FC and 98.8% of FO events were automatically detected. The comparisons show a precision of 4.3 ± 2.9° for knee, 2.9 ± 2.8° for ankle and 4.8 ± 3.4° for hip. Correlation coefficient of all joint angles are higher than 0.97. The proposed system constitutes a new low cost and wearable device allowing to assess stroke patient's locomotion. Indeed, gait event detection accuracy is acceptable since the data are sampled each 14 ms. The joint angles have a high correlation coefficient and the differences between the 2 systems are limited. Its performance and easy use make it a good candidate for pathological gait analysis in ecological conditions.

Determination of the vertical ground reaction forces acting upon individual limbs during healthy and clinical gait

In gait lab, the quantification of the ground reaction forces (GRFs) acting upon individual limbs is required for dynamic analysis. However, using a single force plate, only the resultant GRF acting on both limbs is available.The aims of this study are (a) to develop an algorithm allowing a reliable detection of the front foot contact (FC) and the back foot off (FO) time events when walking on a single plate, (b) to reconstruct the vertical GRFs acting upon each limb during the double contact phase (DC) and (c) to evaluate this reconstruction on healthy and clinical gait trials.For the purpose of the study, 811 force measurements during DC were analyzed based on walking trials from 27 healthy subjects and 88 patients. FC and FO are reliably detected using a novel method based on the distance covered by the centre of pressure. The algorithm for the force reconstruction is a revised version of the approach of Davis and Cavanagh [24]. In order to assess the robustness of the algorithm, we compare the resulting GRFs with the real forces measured with individual force plates. The median of the relative error on force reconstruction is 1.8% for the healthy gait and 2.5% for the clinical gait. The reconstructed and the real GRFs during DC are strongly correlated for both healthy and clinical gait data (R2=0.998 and 0.991, respectively).

Comparison of kinematic and pressure measurement reference methods used in gait event detection

Purpose: A variety of methods have been proposed for detection of initial contact (IC) and foot off (FO) and some comparative analysis is reported in the literature. Pressure measurement insoles and kinematic systems are often part of footwear analysis. Although gait event detection algorithms using these systems have been proposed and evaluated against kinetic data (KN) from force platforms (typically used as a ‘gold standard’), they have not been compared directly. The objective of this work was to undertake this comparison using the same volunteer dataset and test conditions.Methods: Data from 10 healthy adults walking at self-selected normal speed were collected. Two kinematic algorithms (one using a fixed threshold, KM, and the other using high pass-filtering, HPA) and one algorithm using pressure measurement data from an insole (Contact Area Detection, AD) were tested and compared with the detection provided by KN. All data were synchronised and sampled at 200 Hz. Six basic spatio-temporal parameters were also calculated.Results: The absolute mean difference (AMD) in event detection between the three methods and KN was below 25 ms. However, the methods presented tendencies to detect events earlier or later than KN and this influenced the AMD between the methods, which increased to 47 ms for IC detection between HPA and AD. The spatio-temporal parameters showed no statistically significant differences between AD and KM, but differences reached statistical significance between AD and HPA.Conclusion: It is possible to compare gait events and basic spatio-temporal parameters detected using data from pressure measurement insoles and kinematic algorithms; however the kinematic algorithm used will influence the results. Hence the comparison of findings from alternative detection methods is an important issue for which information about the behaviour of the method used is required.

Gait event detection during stair walking using a rate gyroscope.

Gyroscopes have been proposed as sensors for ambulatory gait analysis and functional electrical stimulation systems. These applications often require detection of the initial contact (IC) of the foot with the floor and/or final contact or foot off (FO) from the floor during outdoor walking. Previous investigations have reported the use of a single gyroscope placed on the shank for detection of IC and FO on level ground and incline walking. This paper describes the evaluation of a gyroscope placed on the shank for determination of IC and FO in subjects ascending and descending a set of stairs. Performance was compared with a reference pressure measurement system. The absolute mean difference between the gyroscope and the reference was less than 45 ms for IC and better than 135 ms for FO for both activities. Detection success was over 93%. These results provide preliminary evidence supporting the use of a gyroscope for gait event detection when walking up and down stairs.

Anticipatory postural adjustments contribute to age-related changes in compensatory steps associated with unilateral perturbations

Compensatory steps are essential for preventing falls following perturbations. This study aimed to explore age-related changes in compensatory steps to unilateral perturbations, specifically in terms of whether anticipatory postural adjustments (APAs) play a role in stabilizing lateral balance. Five young and five elderly male adults participated. The split-belt treadmill was used to provide bi- and unilateral perturbations, as forward or backward transitions, applied 10 times in random order. Backward steps evoked by unilateral forward perturbations were evaluated. We measured temporal characteristics, mediolateral (ML) center of mass (COM) motion, and ML step length of compensatory steps. Compensatory steps to unexpected perturbations showed delayed onset of foot-off (FO) and expanded lateral swing length in elderly compared to young subjects. Differences in COM motions and step width arose related to APAs. Elderly subjects showing APAs exhibited no significant differences in ML COM, ML COM velocity, or ML swing length compared to young subjects. However, elderly subjects without APAs showed significant changes toward instability in these parameters. The fact that APAs play a notable role, particularly in the elderly, in stability offers a new insight into preventing falls. However, APAs occurred in 29% of the steps of young and 35% of the steps of elderly subjects. If the occurrence of APAs in elderly people in response to compensatory steps was more frequent, fall risk would be reduced. Further studies, particularly into APA frequency, might contribute to improved intervention to prevent falls.

Gait Event Detection on Level Ground and Incline Walking Using a Rate Gyroscope

Gyroscopes have been proposed as sensors for ambulatory gait analysis and functional electrical stimulation systems. Accurate determination of the Initial Contact of the foot with the floor (IC) and the final contact or Foot Off (FO) on different terrains is important. This paper describes the evaluation of a gyroscope placed on the shank for determination of IC and FO in subjects walking outdoors on level ground, and up and down an incline. Performance was compared with a reference pressure measurement system. The mean difference between the gyroscope and the reference was less than −25 ms for IC and less than 75 ms for FO for all terrains. Detection success was over 98%. These results provide preliminary evidence supporting the use of the gyroscope for gait event detection on inclines as well as level walking.

Online gait event detection using a large force platform embedded in a treadmill

Gait research and clinical gait training may benefit from movement-dependent event control, that is, technical applications in which events such as obstacle appearance or visual/acoustic cueing are (co)determined online on the basis of current gait properties. A prerequisite for successful gait-dependent event control is accurate online detection of gait events such as foot contact (FC) and foot off (FO). The objective of the present study was to assess the feasibility of online FC and FO detection using a single large force platform embedded in a treadmill. Center-of-pressure, total force output and kinematic data were recorded simultaneously in 12 healthy participants. Online FC and FO estimates and spatial and temporal gait parameters estimated from the force platform data—i.e., center-of-pressure profiles—were compared to offline kinematic counterparts, which served as the gold standard. Good correspondence was achieved between online FC detections using center-of-pressure profiles and those derived offline from kinematic data, whereas FO was detected 31 ms too late. A good relative and absolute agreement was achieved for both spatial and temporal gait parameters, which was improved further by applying more fine-grained FO estimation procedures using characteristic local minima in the total force output time series. These positive results suggest that the proposed system for gait-dependent event control may be successfully implemented in gait research as well as gait interventions in clinical practice.

Detection of gait events using an F-Scan in-shoe pressure measurement system

A portable system capable of accurate detection of initial contact (IC) and foot off (FO) without adding encumbrance to the subject would be extremely useful in many gait analysis applications. Force platforms represent the gold standard method for determining these events and other methods including foot switches and kinematic data have also been proposed. These approaches, however, present limitations in terms of the number of steps that can be analysed per trial, the portability for outdoor measurements or the information needed beforehand. The purpose of this study was to evaluate the F-Scan ® Mobile pressure measurement system when detecting IC and FO. Two methods were used, one was the force detection (FD) in-built algorithm used by F-Scan software and a new area detection (AD) method using the loaded area during the gait cycle. Both methods were tested in ten healthy adults and compared with the detection provided by a kinetic detection (KT) algorithm. The absolute mean differences between KT and FD were (mean ± standard deviation) 42 ± 11 ms for IC and 37 ± 11 ms for FO. The absolute mean differences between KT and AD were 22 ± 9 ms for IC and 10 ± 4 ms for FO. The AD method remained closer to KT detection for all subjects providing sufficiently accurate detection of both events and presenting advantages in terms of portability, number of steps analysed per trial and practicality as to make it a system of choice for gait event detection.

Ambulation After Incomplete Spinal Cord Injury With EMG-Triggered Functional Electrical Stimulation

Individuals with incomplete spinal cord injury (iSCI) retain some control of the partially paralyzed muscles, necessitating careful integration of functional electrical stimulation (FES) with intact motor function. In this communication, the volitional surface electromyogram (sEMG) from partially paralyzed muscle was used to detect the intent to step in an iSCI volunteer. The classifier was able to trigger the FES-assisted swing phase with a false positive rate less than 1% and true positive rate of 82% for left foot-off (FO) and 83% for right FO over 110 steps taken during three testing sessions spread over a week.

Validity of an automated gait event detection algorithm in children with cerebral palsy and non-impaired children

The signal obtained from calculating the ratio between he toe marker resultant velocity and its anteroposterior (AP) omponent in the sagittal plane can be used to detect foot ontact (FC) and foot off (FO) events using an automated lgorithm. This technique can be potentially used as an alterate method for the detection gait events when data from orce plates or foot-switches is limited or not available.