Clinical Gait Data Processing and the Assessment of Gait Abnormalities within a
Neurorehabilitation Program
submitted in partial fulfillment for the degree of master of science Callum Hsiao
12850748
master information studies data science
faculty of science university of amsterdam
2020-11-26
Internal Supervisor External Supervisor UvA Examiner Title, Name Dr Erik Bekkers Ruud van der Veen Dr Frank Nack Affiliation UvA, AMLab Daan Theeuwes Centrum UvA
Clinical Gait Data Processing and the Assessment of Gait
Abnormalities within a Neurorehabilitation Program
Callum Hsiao
clhxiao@gmail.com University of Amsterdam
ABSTRACT
Gait analysis(GA) is a powerful tool for analysing locomotion biome-chanics, and it is particularly useful in the context of rehabilitation for determining appropriate treatments and to evaluate the effective-ness of interventions. This study approaches clinical gait analysis within the context of a neurorehabilitation program using data collected form a rehabilitation treadmill. A streamlined approach to assess gait abnormalities for the purpose of clinical evaluation was developed and validated. Using an automated rule based method and vertical ground reaction force data, raw gait information is segmented and processed into gait cycles for feature extraction. In order to objectify clinical impression and document changes in gait pattern, four normalcy index for overall gait, hip joints, knee joints and trunk flexion-extension were obtained using a method based on principal component analysis. The results demonstrate a potential for the normalcy index to be used to quantify abnor-mal gait based on its degree of deviation from norabnor-mality, and may be adapted within future research as a robust metric to evaluate rehabilitation progress and aid with prognosis modeling.
KEYWORDS
Gait Analysis, Clinical Assessment, Normalcy Index, Kinematics, Biomechanics, Neurorehabilitation, Acquired Brian Injury
1
INTRODUCTION
Gait is one of the most complex motor skills in humans, relying on the coordination of multiple systems, including the central nervous system, peripheral nervous system and the musculoskeletal system in order to produce a stable gait and consistent walking pattern [21][28]. Acquired brain injury (ABI), one of the most common cause of death or disability among youths and young adults, can result from events with an external or internal cause and lead to various functional impairments among which gait dysfunction is common [33][12]. Since impairments that contribute to gait disor-ders can vary considerably in its severity and type, the way each individual responds to therapy depends on a variety of variables [45]. As a result, gait training is often a significant focus of neu-rorehabilitation programs for patients with severe acquired brain injury [47].
Gait analysis is the systematic study and analysis of human loco-motion, and is often carried out in clinical practices to identify gait impairments [6]. It is a powerful assessment tool for surgical deci-sion making, post operative follow up, and obtaining information that can help establish and evaluate the level of functional limita-tion of gait due to pathology [8] [20] [6]. Tradilimita-tionally, gait analysis is performed by therapists using a qualitative approach based on direct visual observations [30]. However, while observational gait analysis plays an important role in clinical decision making and
evaluation, such approach can be also be time consuming and non-objective [2]. As such, following the advancement of biotechnology, increased importance have been placed on quantitative gait analy-sis. Through utilising joint angle kinematics and kinetics, ground reaction force and dynamic electromyography measurements that can be collected using modern technology, these automatic systems constitute an objective technique that is often more precise and cost effective [30][11][15]. Additionally, through the employment of modern data science methods, manipulation and insight extraction from large and more complex datasets can also be accomplished, further advancing the study of the biomechanics of human gait [14].
This project is proposed by the Daan Theeuwes Center (DTC) for Intensive Neurorehabilitation in Woerden, which specialises in providing intensive and prolonged multidisciplinary treatments to young people with severe acquired brain injury. The quality care program concerned aims at improving the care offered based on structured clinical data collection including a comprehensive profile of demographic and clinical characteristics, as well as prospective rehabilitation outcome measurements [36]. For a preliminary anal-ysis, the targeted data for exploration is the gait measurements collected using the machines newly introduced to the clinic as part of the physical rehabilitation therapy. Specifically, the inter-pretation and analysis of the data collected by the rehabilitation treadmill WalkerView 3.0 from TechnoBody Italy is of interest to the therapists at the DTC.
The broad goal of this assignment is to explore the available data, and evaluate the extent to which it can be of value to the treatment at the Daan Theeuwes Centre. This project therefore aims to develop a processing pipeline to streamline the assessment process of affected gait in the future and inform clinical decision making and direct intervention programs [46]. The work should set a foundation for the systematic data collection that will aid future research for the development of prediction models that can help with rehabilitation prognosis. With this aim, this project will attempt to answer the following research question: How can the structured data collected by the WalkerView 3.0 treadmill machine be used to aid the assessment of abnormal gait within a neurorehabilita-tion program through the use of quantitative analysis? To answer the research question, the main problem is divided into the following sub-questions, characterised by the process of clinical gait data analysis:
• How can gait cycle extraction for gait analysis using kinetic and kinematic gait data collected from the WalkerView be automated?
• What relevant gait features can be obtained from the Walk-erView data and how can it be interpreted for clinical assess-ment through quantifiable measures?
• To what extent can the extracted gait features and summary measures help assess common pathological gait patterns in patients with acquired brain injury and characterise abnor-mal gait in the evaluation of rehabilitative treatment pro-gresses? What value can it add to the overall rehabilitation program at the DTC?
This paper is divided into 7 sections. Following the above intro-duction of the research context and project aims, section 2 and 3 will address the relevant background knowledge regarding clinical gait analysis and existing research within the field. Section 4 will then explain the method of data processing and analysis employed for this project. In section 5, the results of the quantitative data analysis as well as the outcomes of said experiments will be pre-sented. The final sections will then discuss the obtained results and assess its relevance in the wider context of the intensive care pro-gram at the Daan Theeuwes Center. Addtionally, the added value of integrating the WalkerView 3.0 as part of the neurorehabilitation treatment program will be discussed, followed by an acknowledge-ment of the limitations of this study and a conclusion providing recommendations for further research.
2
BACKGROUND
This section provides the background knowledge regarding ac-quired brain injuries and the fundamentals of clinical gait analysis in which the analytical study is based upon.
2.1
Acquired Brain Injuries
Acquired brain injury (ABI) is any injury to the brain that occurs postnatal, and is not hereditary, congenital or degenerative [33]. The resulting brain injuries can of traumatic or non-traumatic origin, and the subsequent impairment could be either temporary or permanent [41] [32]. The neurological consequences of an ABI can severely affect the cognitive function of the victim through compromising the physical integrity or function of one or more areas of the brain and lead to partial or total disability [19]. In addition to causing harmful clinical, social and economic effects, the injured individual’s functional and psychosocial recovery prospects can also be heavily affected [19].
2.2
Motion Analysis and Biomechanics
Biomechanics is the application of Newtonian mechanics to the study of the neuromuscular system [40]. Most commonly used in orthopaedics and the characterisation of function and dysfunction of the muscular skeletal system, its principals are fundamental to the study of human gait [40].
2.2.1 Gait Cycles, Gait Phases and Gait Events. Human gait com-prises a periodic phenomenon in which motions in both legs pro-duces a series of repetitive patterns called gait cycles that allows an individual to move from one point to another [26]. The gait cycle is defined as the period from one event of one foot to the following occurrence of the same event with the same foot, gener-ally designated by the initial contact (also known as heel-contact, heel-strike or rearfoot strike), in which a foot contacts the ground heel-first during the foot strike phase of a walking cycle [16]. The gait cycle consists of one stride length further divided into a stance
and swing phase, which is a period of weight-bearing and interval of self-advancement respectively [16]. Each phase takes up approx-imately 60% and 40% of the gait cycle respectively and are classified by a cut-off gait event toe-off (also known as foot off), such that the interval between the toe-off point and heel-strike for a foot represents the swing phase while the stance phase occurs between heel-strike and toe-off [26]. The two phases can then further be di-vided into seven sub-phases, including: Load Response, Mid Stance, Terminal Stance, Pre-Swing, Initial Swing, Mid-Swing and Terminal Swing [34] [37]. The beginning and end of each of these gait phases are marked by specific gait events as seen in figure 1.
2.2.2 Spatio-Temporal Parameters of the Gait Cycle. Spatial and temporal parameters are generally recognised as key metrics for characterising gait, wherein objective measures of such parameters allow for the characterisation of functional gait performance [5][9]. The identification of specific gait events are required in order to compute the required spatio-temporal parameters. At the very least, the initial contact and toe off moment must be identified in order to identify the steps and strides and their parameters within a given set of gait data.
Spatial gait parameters generally refer to step and stride length, which can be defined from the distance covered between two con-secutive Initial Contacts [5]. Meanwhile, the most common used temporal parameters for gait analysis include stride and step du-ration and cadence. Each gait cycle generally lasts the dudu-ration of around one second, also called stride time, and can further be divided into double support stances and single support stances [40]. Double support is the period of time when both feet are in contact with the ground which occurs at the beginning and end of stance phase [16]. During double support, the weight is transferred from one foot to the other [40]. Single support is the period of time when only one foot is in contact with the ground [16]. During a normal active stride, it is equal to the swing phase of the other limb and describes when the center of mass of the body passes over the foot in preparation for shifting to the other limb [40]. Furthermore, the duration of a temporal parameter can be characterised using unit in seconds, or as a percentage of a gait cycle.
2.2.3 Body Segments and Joint Kinematics. In order to analyse hu-man gait, the assumption is that each body segments involved in the motion can be modeled as rigid bodies, where the position and motion of the underlying skeleton can be tracked using a biome-chanical model in 3D space[40]. Most often used for such analysis is the Cartesian coordinate system, which consists of the anterior-posterior, medial-lateral, and longitudinal axis, corresponding to the frontal, sagittal and tranverse planes respectively [1]. Furthermore, motion at the pelvis, hip, knee and joint ankles allow the analysis of temporal and stride events of the gait by recording the biome-chanical angular motion of the joints, including flexion-extension, internal-external rotation and abduction-adduction [20] [1]. Exam-ples of normative gait data can be seen in figure 5 in Appendix A.
2.2.4 Motion on the Sagittal Plane During the Gait Cycle. Due to the constraint of the available data for this study, the focus was placed on hip and knee joint kinematic data on the sagittal plane. Critical events within a gait cycle including these parameters are
Figure 1: Gait Cycle and Gait Events
described here and can be seen summarised in figure 6 in Appendix A.
Hip Flexion-Extension During Gait. Only a single arc of hip ex-tension and flexion occurs in each gait cycle, and during the initial contact of a foot, the corresponding hip joint is generally in around 30 degrees of flexion [31]. The hip joint motion then progressively extends into hyperextension, typically peaking at around 10 degrees close to the toe-off event half way through the gait cycle, as flexion begins again in terminal double stance and continues through most of the swing phase, peaking at around 85% of the gait cycle where maximum hip flexion is achieved [31][27].
Knee Flexion-Extension During Gait. The knee joint goes through two phases of flexion and extension in each gait cycle unlike the hip. During initial contact, the knee begins in full extension (or flexed roughly at 5 degrees), the rapidly flexes during the loading response, reaching a first peak usually at around 15 degrees, then progressively extends to neutral until it begins to flex again with the onset of the double stance [31]. This flexing action then continues into the swing phase until it reaches a maximum flexion to roughly 60 to 70 degrees at the beginning of mid-swing (73% of the gait cycle) before extension is resumed until the next gait cycle [31]. 2.2.5 Ground Reaction Forces and COG. Ground reaction force (GRF) is the only other force acting on the body while walking aside from gravitational attraction if negating wind, and can be measured using by a pressure sensitive force plate on a flat surface in which the gait measurement is conducted [40]. Output data of the measurements can be processed into ground reaction force vector components that represent vertical load, shear loads, torque about the vertical axis, and body center of pressure location [20]. During gait analysis, GRF can reveal minor changes of the gait pattern as well as shifts in the centre of gravity, and can therefore validate the state of disorder of a subject’s movement [48].
2.2.6 Kinetic Studies. With information on body segment motions gathered from kinematic analysis, force data gathered from either a force plate or force dynamometer can then be applied to it to calculate the forces causing motion using Newton’s second law. Through application of equilibrium equations, the joint moments can be computed and normalised and expressed as a percent of body weight times leg length. Furthermore, if the moment and joint are known, joint power can be calculated and incorporated into gait
analysis to provide insight into subtle functional musculoskeletal adaptations [20].
2.3
Gait analysis
Human gait analysis, being a critical component of the analysis of individual locomotion function, has seen extensive research and a wide range of applications. The two main topics of general interest are gait identification and gait analysis for clinical applications.
2.3.1 Gait Identification. Gait identification refers to the recogni-tion of individuals based on their gait patterns and is generally used in biometric identification, healthcare monitoring or surveillance [2][22]. Gait identification employs computer imaging technology and are commonly based on deep learning models.
2.3.2 Clinical Gait Analysis. On the other hand, clinical analysis of gait aims to evaluate locomotion movements for the purpose of gaining insight on human movement patterns corresponding to different gait pathology, which can support diagnoses and therapy considerations among other clinical applications [1][29]. The main aim for clinical gait analysis, in addition to understanding general human movement, is to assess and diagnose pathological gait. The study of neuromuscular disorders and injuries is an area where gait analysis offers great value, providing a non-invasive technique to evaluate the effects of neurological impairments on gait [13]. Gait analysis consists of the collection of biomechanical data and the subsequent processing and manipulation of it. 3 dimensional gait analysis (3DGA) is the standard method for quantifying biome-chanical abnormalities during gait, and high level biomebiome-chanical data generated from 3D gait analysis have been shown to positively affect clinical decision making and treatment outcomes in clinical applications such as the analysis of neurological conditions such as cerebral palsy, strike, and Parkinson’s disease [44].
The core component of most contemporary gait analysis is the measurement of joint kinematics and kinetics, where temporal and spatial characteristics of the gait cycle is recorded [3] [35]. Other measurements regularly made include electromyography, oxygen consumption and foot pressure [3]. Generally, gait data appear as temporal waveforms representing specific joint measures throughout the gait cycle [13].
3
RELATED WORK
Given the background knowledge regarding gait biomechanics and the function of clinical gait analysis, this section provides a brief overview of the existing research concerning the use of data science techniques to process and analyse clinical data, and how clinical gait analysis are applied within rehabilitation programs.
3.1
Gait Segmentation
After collecting gait data from recording different parts of the body, important gait features are usually extracted from individual gait cycles for assessment and analysis. Generally, when analysing gait, the accurate and consistent segmentation of gait cycles from a data set containing multiple cycles is essential, as is the detection of the main gait phases of stance and swing [24] [25].
3.1.1 Automated Gait Event Identification. In most studies, algo-rithmic methods to detect the gait events that mark the transition between each gait phases were employed, and depending on the research, the aim may be to identify multiple gait sub-phases, or just the two main ones stance and swing. However, since there are no standard methods to record gait data, it is likely every research present in the available literature is working with a differently structured dataset. Nonetheless, gait segmentation are generally performed using similar strategies such as identifying peaks in signals that correspond to specific gait events such as initial con-tact and toe off, and often utilise ground force data, foot switch recordings, kinematic data or a combination of two or more.
Ghoussayni et al. in [18] compared and validated a kinematic based algorithm used in the detection of heel contact, heel rise, toe contact, and toe off. 3D marker coordinate data was filtered wherein marker velocities in the sagittal plane were then calculated to determine the timing of the gait events using empirically set thresholds [18]. Likewise Khan and Badii [26] uses hip joints motion data and a rule based algorithm to partition gait data into seven subphases. Jiang et al. [24] proposed a method based on a peak detection approach that can process different type of gait signals and has a enhanced ability to segment gait cycles by eliminating the false peaks and interpolating the missing peaks, which was demonstrated to be effective through testing on data of patient’s diagnosed with Parkinson’s disease.
3.2
Gait Features and Summary Measures
After identifying gait cycles the next step of the analysis is to define and extract parameters and features from the sample signal as descriptors of discrete instants or events of the gait pattern [13]. Feature extraction are generally approached locally or globally. The local method in the case of gait analysis consists of describing the biomechanical data based on some specific points extracted from the acquired temporal waveform, including features such as summary statistics or parametrization involving measures on a single biomechanical gait data.
3.2.1 Principal Component Analysis. The usefulness of principal component analysis for gait feature extraction is often asserted due to its effectiveness at dimensionality reduction and feasibility for visual interpretation. While retaining the temporal characteristics, PCA can summarise information contained in the gait cycle and
reduce information contained in the biomechanical gait waveforms into a small number of principal components [7]. Moreover, the features elicited by PCA also often agree with the most clinically relevant features [1]. As such, PCA is often used as a data reduction tool as well as preliminary step for further analysis to determine differences between patient and control groups, as demonstrated in Deluzio and Astephen’s study on knee osteoarthritis features identification [13] and can also be seen used in various studies of pathological gait or gait patterns such as analysis of gait varia-tions in stroke patients by Boudarham et al.[7] and classification of neurological disorders of gait by Pradhan et al. [35].
3.2.2 Gait Summary Measures. While gait analysis provides an effective tool for evaluating and quantifying the effects of a clinical intervention or other treatment on a patient’s gait, objectively quan-tifying the degree to which a patient’s gait has improved following an intervention remains difficult [8][39]. Over the last decade the need for a concise index to summarise gait characteristics or to measure the ‘quality’ of a particular gait pattern has been raised as typical 3D-GA evaluation produces a vast amount of data, and despite its objectivity, it can be difficult to interpret [38]. Several gait summary measures used in conjunction with 3DGA have been proposed and researched for their possibility to objectify clinical impression and quantify the degree of gait deviation from normal, stratify the severity of pathology and document changes in gait over time [8]. Notable ones include the gait normalcy index, hip flexor index, gait deviation index and gait profile score among others [8].
4
METHODS
Following the establishment of the context of this research, the methods employed to answer the research questions are explained in this section. Specifically, this research aims to address the practi-cal needs at the Daan Theeuwes Centrum, and takes a focus on the development of a data processing pipeline and subsequent experi-ments.
4.1
Data Description
First, the data used in this study collected by the therapists at DTC using the WalkerView 3.0 treadmill is described.
4.1.1 Sample population. The population of the admitted patients at the DTC consists of young adults between the ages 16 and 35 who have severe acquired brain injuries and seeks an intensive rehabilitation treatment. Around 46% of the patients were admitted as a result of traumatic brain injury, while 36% suffered a stroke and the remaining 18% from other non-traumatic causes1. The patient population that are participants for this preliminary study consists of 21 subjects. All are currently undergoing treatment and have had their gait performance measured by the rehabilitation treadmill 1 to 4 times over the span of a few months between February to August 2020. Additionally, to provide healthy control samples for asymptomatic gait, 15 healthy subjects from the DTC were measured to be used as preliminary reference samples for this study.
1Information given by the DTC
4.1.2 Instrumentation. The WalkerView 3.0 developed by Techno-body Italy was used to collect the gait data sample. The treadmill is equipped with a sensorized belt with eight load cells and allows the assessment of stance during ambulation and corrects dynamic parameters in real time [42]. The system is also equipped with a 3D camera, which can receive instant and objective dynamic image of the posture during training and display it in real time on the Full HD screen of the system. The 3D camera, using two optics, a classic one for HD 2D shootings and an infrared one allows for the analysis of a subject’s gait biomechanics as it can reliably capture any single angular movement of the person walking on the treadmill [43]. 4.1.3 Data Acquisition. Gait data were collected by having the subjects complete a 2 minute standard test on the treadmill in which the WalkerView system will store into its internal operating system. Structured data in the form of a csv file are saved to the system while it also generates a gait analysis report (see Appendix B). The report presents basic gait parameters such as the range of motion, maximum and minimum value of the trunk, hip and knee joints. Load symmetry as well as standard temporal parameters such as the average step cycle time, step length, contact time and vertical COG displacement are presented as well. The report also includes graphs of the measured waveforms as well as the normalised and segmented gait cycles, which is overlayed with a comparison of the entire detected biomechanical waveform and a normative range reference (Appendix B fig 9).
At the current stage, however, the gait analysis report provided by the WalkerView is of ambiguous value to the therapists. Interest in the value of the quantitative data measured by the machine is present, however it is yet unclear what notable information can be obtained through the machine. With that, the following sections seek to explore and process the raw data stored in the csv files and develop a processing pipeline where specific analysis regarding gait abnormalities can be conducted through the use of quantitative methods.
Examining the structured data stored in the csv file, it can be observed that the WalkerView records measurements of ground force reaction information, kinematic information of the trunk, hip and knees on the sagittal plane, the displacement of the center of gravity and the left and right load. In figure 7 from Appendix B, visualisations of the temporal waveforms of the measured gait data of a random healthy control subject can be seen. Without any preprocessing, the ground force reaction data collected by the 8 load cells on the treadmill do not line up temporally with the kinematic data due to different sampling frequencies.
4.2
Gait Partition
With the time series data collected, gait cycles can be extracted through the identification of key gait events for the quantitative analysis of key gait parameters. For this preliminary analysis, a rule based gait segmentation algorithm utilising hip joints angular data was applied as hip flexion and extension are one of the most important muscular activities in human locomotion [10]. It can be observed in figure 7 that hip joint motion in the sagittal plane moves in a relative periodic fashion. It is implied that the maximum and minimum values for hip joint sagittal dimension angular motion indicate the flexion and extension extremes of the corresponding leg
respectively in a gait cycle [26]. Thus, a kinematic based algorithm that exclusively processes hip joint data to partition gait phases was employed in order to obtain gait cycles for further analysis.
In addition to extracting gait cycles, identifying gait events within each gait cycle is also an important procedure when analysing gait. For this, ground reaction force data was used in order to deter-mine the point in which each key gait event occurs, partitioning the gait cycles into the following gait events: Loading Response (LR), Mid Stance (MSt), Terminal Stance(TSt), Pre Swing(PSw), Initial Swing (ISw), Mid Swing (MSt), Terminal Swing(TSt).
A visual representation of the algorithm can be seen in figure 2. The output of each algorithm is a list of arrays of the starting indices of every partitioned gait sub-phase. The results were entered into another function that obtains the start and end point of every gait phase. First, gait segmentation was performed using smoothed hip kinematic data as input, after which each gait cycle was extended or compressed in time to yield a normalised gait cycle of 101 data points and expressed as a function of a unit 100% cycle length, irrespective of the actual time for a stride for further analysis and the extraction of gait feature variables. The ground reaction force data for each processed gait cycle were then used as input for the second algorithm in order to obtain the time each gait event occurs within a given gait cycle. Specifically, the left and right side average GRF obtained from averaging the GRF data from cell 1,3,5,7 and 2,4,6,8 were used to determine the end point of the Mid Stance, Terminal Stance and Initial Swing of each gait cycle, and information from cell 3 and cell 4 in particular were used to find the point of Toe Off (end of Pre Swing), Loading Response and Mid Swing. The results obtained were compared to normative reference data in order to determine the accuracy of the algorithms, additionally, abnormal instances of identified gait events were counted.
Figure 2: Hip Joint Based Gait Segmentation Algorithm
4.3
Gait Analysis
From the gait cycles obtained as well as the indices of each gait sub-phases, key gait events were used to calculate spatio-temporal and sagittal kinematic variables for the gait cycles of interest. Clinically relevant gait features that can be obtained through the WalkerView data can be seen summarised in table 6 in Appendix D. Summary statistics were then generated for all the key gait variables and
compared to the asymptomatic reference using a t test where ob-servations can be made to identify incidences of gait abnormality following acquired brain injuries. Additionally, individual results for the patient sample were compared with the 95% confidence intervals calculated for the corresponding variable for the healthy control sample, and values within ±2 standard deviations from the mean were categorised as normal [46]. For the assessment of indi-vidual gait for clinical evaluation, detailed gait profiles that provide more accurate and objective results compared to observational gait analysis were generated, adding to the gait analysis report provided by WalkerView. This allows for the evaluation of specific gait char-acteristics as part of the assessment of clinical interventions, and can help identify potential gait problems.
4.3.1 Normalcy Index. To accurately evaluate the extent of gait deviations from normal gait, or to assess the changes in a gait resulting from a specific treatment, it is important to consider not only how each feature of the gait pattern has changed but also how the relationship between the features changed. To evaluate whether a specific gait variable is normal, abnormal, or improved following treatment, the natural correlation that exists between gait variables must be determined [39]. For this reason multivariate statistical techniques were used to develop a measure of how closely an individual gait pattern approaches normal. This ‘closeness’ is referred to as the normalcy index, which is the measure of the distance between the set of discrete variables describing a patient’s gait pattern and the average of those variables in persons with no gait abnormalities [38].
Using a proposed method based on principal component analysis, the normalcy index quantifying gait deviations from the average of normal gait was obtained [38]. First, a set of 𝑁 discrete gait variables that are correlated with the aspects of gait that the normalcy index will be describing were selected, and represented by 𝑥𝑗, 𝑗 = 1, 𝑁 . The process to obtain the normalcy index was as follows:
(1) Calculate the mean (𝜇𝑗) and standard deviation (𝜎𝑗) of 𝑥𝑗 measured on M normal subjects.
(2) Standardise the data, defined by 𝑧𝑗 = (𝑥𝑗− 𝜇𝑗)/𝜎𝑗
(3) Calculate the covariance matrix (𝐶𝑖 𝑗) for the 𝑁 standardised discrete variables. (𝑖 & 𝑗 = 1, 𝑁 )
(4) Calculate the eigenvalue-eigenvector pairs (𝜆𝑖− 𝑒𝑖) for (𝐶𝑖 𝑗), a subset of the eigenvectors will be used as basis vectors (5) To obtain the normalcy index for a subject, let it be be
repre-sented by the same N discrete variables ˜𝑥𝑗, 𝑗 = 1, 𝑁 (6) Standardise ˜𝑥𝑗using the mean (𝜇𝑗) and standard deviation
(𝜎𝑗) from the normal subjects ˜𝑧𝑗 = (𝑥˜𝑗− 𝜇𝑗)/𝜎𝑗
(7) Project ˜𝑧𝑗 onto the basis vectors obtained from 𝑒𝑖and stan-dardise it by dividing it by the square root of 𝜆𝑖 to obtain
˜ 𝑦𝑖
(8) Find the square of the Euclidean length of ˜𝑦𝑖 for a given subject 𝑑= 𝑁 Õ 𝑖=1 ˜ 𝑦2 𝑖
The number 𝑑 represents the square of the distance of an individual subject’s data from the normal mean in the new uncorrelated coordinate system and is defined as the nor-malcy index [38].
For this study, in addition to defining and exploring a general gait normalcy index based on gait variables that are generally considered to be important and can be obtained through the available data, specific joint normalcy indices that aim to evaluate gait pathologies at the joint level were also defined. Parameters were selected from the WalkerView data referencing previous studies, and the extracted gait features and joint parameters are can be seen summarised in in tables 7, 8, 9 and 10 in Appendix E [39][49].
To evaluate of the selected variables, the consequences of the choices for each index was recalculated leaving out one variable each time. Correlation coefficients between the index with all 𝑁 parameters and index with one excluded parameter would indicate whether a variable seem to dominate the final value [38]. As for the validation of the obtained normalcy indices in clinical cases, the differences between the healthy subjects and patients were compared.
4.4
Case Studies
To assess how well the normalcy index can assist with the evalua-tion of a subject’s gait performance in a clinical setting, case studies were conducted where six patients currently receiving treatment at the DTC were selected based on data availability and separated into two groups based on observational judgment of the severity of their gait impairment. Gait analysis results from patient data collected over the past few months were visualised and compared to their clinical assessment data from the DTC.
5
RESULTS
5.1
Automated Gait Segmentation
Examples of gait segmentation results can be seen table 1 and figure 10 in appendix C. In an average normative gait cycle, Load Response usually ends at around 12% of the gait cycle, while Heel Rise occurs at around 30%, terminating the Mid Stance phase. Furthermore, Terminal Stance ends at around 50% following Heel Rise, and Toe Off occurs at around 62% which marks the end of the stance phase of a gait cycle. Toe Off is then followed by Initial Swing which ends at roughly 75% of the gait cycle, and lastly Mid Swing ends at 85%, leading to the final Terminal Swing phase [23].
For normal gait behaviour, gait event partition results as ob-tained from ground reaction force data appears to be reliable and consistent, while more challenges are present in the partitioning of abnormal gait. This is especially the case for determining each gait subphases due to the more irregular vertical ground reaction force data (see fig 10 in appendix C), which can potentially affect the extraction of certain gait features. Results in table 1 presents the mean value where each gait subphase ends within a gait cycle. Individual gait cycles for each subject where the results are deviate from the mean value by more than 2.5 standard deviations were deemed as abnormal. From the results it can be seen that around 9% of the gait cycles for each measured subjects have abnormal gait event occurrences as partitioned using the proposed algorithm,
Table 1: Gait Phase Duration
Stance Phase Swing Phase
Loading Response Mid Stance
(Heel Rise) Terminal Stance
Pre Swing
(Toe Off) Initial Swing Mid Swing % Abnormal Healthy L 12.05 ± 2.03 27.64 ± 1.36 45.25 ± 4.82 64.15 ± 1.64 77.53 ± 4.14 87.50 ± 5.92 9.45% ± 3.91%
R 10.25 ± 2.26 24.15 ± 2.99 42.05 ± 5.57 60.93 ± 2.42 80.95 ± 2.90 89.39 ± 4.09 7.72% ± 4.40% Patient L 15.05 ± 6.41 27.34 ± 5.43 43.67 ± 6.02 64.04 ± 6.10 77.67 ± 3.58 82.89 ± 4.49 9.64% ± 5.16% R 14.94 ± 6.07 26.26 ± 4.77 41.56 ± 5.72 63.02 ± 5.51 80.90 ± 3.32 88.95 ± 5.11 9.43% ± 5.09%
with patient subjects on average having more abnormal gait cycles compared to healthy subjects.
5.2
Quantitative Gait Analysis
The summary statistics of the healthy control group and the patient data samples are outlined in table 2, while the joint kinematic data in the sagittal plane are summarised in in table 3. Moreover, the frequency for the classification of abnormal gait variables that falls outside of 95% CI can be seen in table 4. As can be observed from the results, the classification of abnormality varies depending on the variable. The patient sample was significantly different from the healthy control samples for 7 out out of 20 variables, and more than a quarter of the samples was classified as abnormal (table 4). Notably, the spatio temporal variables analysed appear to deviate more significantly from the healthy control values, and it can be observed that the patient population generally adopted a slower walking speed as reflected in the decreased cadence and lower stride time. Patients also tend to have a greater stance phase and significantly increased double support duration, with a notable portion of subjects having increased knee flexion at initial contact but decreased knee flexion overall, especially for its range of motion during the swing phase.
5.3
Normalcy Index
The correlation between index calculated with all parameters and one excluded parameters can be seen in tables 7, 8, 9 and 10 in Appendix E along with the specific chosen variables for this analysis. According to the correlation coefficient calculated, it appears that the normalcy index is for the most part rather sensitive to the precise composition of gait variable combinations. The hip joint index and knee joint index shows a slight exception, with the correlation values being relatively closer to 0.90 compared to the general gait normalcy index and trunk index.
Next, in figure 3 and table 5 the normalcy index of healthy and pathological patient subject group are presented. The mean in-dex for the patients are notably higher than that of the healthy control group and showcases a statistically significant difference. This result shows that the proposed gait normalcy index and joint normalcy index was able to distinguish between normal and abnor-mal gait based on the chosen parameters, where patients with a more dysfunctional gait obtaining a higher score in general, with a few notable outliers receiving a significantly higher score (see also figure 13 in Appendix E).
5.3.1 Case Studies. To evaluate the relevance and viability of the normalcy index, a few case studies was conducted. Six patients were
Figure 3: Normalcy Index
chosen based on an observational assessment of their recorded gait performance from the archived WalkerView video data and divided into two groups. Group A were patients who have received a higher Functional Ambulation Category (FAC) score during their initial report and can perform a 2 minute test on WalkerView indepen-dently, while Group B were patients who have more severe gait dysfunction and had to rely walking aid throughout their rehabil-itation. Each selected patients has at least three WalkerView test data available, measured over the period of at least three and a half months.
From information obtained from the earliest available measure-ment for the subjects, the normalcy index obtained from group A were 0.84, 76.22 and 1.5 for overall gait; 16.46, 74.35, 7.77 for hip; 13.25, 246.11, 76.46 for knee and 15.51, 21.38, 25.5 for trunk. For group B the normalcy index values were 50.08, 15.54 and 138.84 for overall gait; 154.28, 123.04, 228.39 for hip; 144.09, 468.66, 204.23 for knee and finally, 348.16, 71.43 and 204.29 for trunk. Significant difference is noticeable between the two groups, as patients with more drastically abnormal gait analysis results receiving higher normalcy index scores as expected. For some subjects, however, it is possible to score low on the normalcy index, indicating a small deviation from an average normal gait pattern, while scoring high on the other indexes. Through observing the difference between higher and lower gait index scores, it appears to mainly evaluate the periodic consistency within the entire gait pattern.
Figure 4 is an example of the gait evaluation report for one of the case studies. The report gives information on the normalcy index, and visualises the gait cycle for the hips, knees, trunk on the sagittal plane as well as the lateral COM displacement. A segment of the raw data is also visualised, with the normalcy index labeled on top
Table 2: Spatio Temporal Variables
Patient (n = 21) Healthy (n=15)
Gait Variable Mean ± SD Range Mean ± SD Range 𝑝 Cadence 108.09 ± 21.31 44 - 134 127.67 ± 12.02 98 - 156 0.003 Stance Duration (s) 0.71 ± 0.25 0.48 - 1.67 0.56 ± 0.06 0.40 - 0.73 0.025 Initial Double Support (s) 0.17 ± 0.08 0.06 - 0.42 0.10 ± 0.02 0.05 - 1.15 0.002 Single Support (s) 0.70 ± 0.19 0.38 - 1.46 0.63 ± 0.11 0.39 - 0.82 0.175 Terminal Double Support (s) 0.25 ± 0.18 0.02 - 0.88 0.17 ± 0.05 0.13 - 0.34 0.076 Stride Time 1.12 ± 0.36 0.83 - 2.50 0.90 ± 0.10 0.66 - 1.15 0.031
Table 3: Kinematic Variables
Patient (n = 21) Healthy (n=15)
Gait Variable Mean ± SD Range Mean ± SD Range 𝑝 Trunk Flexion (deg) 4.17 ± 1.46 2.62 - 8.28 3.40 ± 0.70 2.39 - 4.84 0.066 Trunk Lateral Flexion (deg) 4.04 ± 1.76 1.87 - 8.55 2.97 ± 0.97 1.80 - 4.55 0.040 Hip ROM (deg) 43.29 ± 7.28 30.06 - 54.52 46.09 ± 5.73 33.75 - 53.95 0.224 Hip Flexion at Initial Contact (deg) 24.03 ± 5.20 11.58 - 31.59 26.16 ± 3.59 18.57 - 32.16 0.180 Hip Flexion at Heel Rise (deg) 1.05 ± 3.91 -7.58 - 7.99 2.18 ± 3.17 -2.91 - 7.07 0.362 Hip Peak Flexion (deg) 33.55 ± 7.62 21.77 - 55.21 32.23 ± 7.71 24.09 - 56.75 0.613 Knee ROM during Stance (deg) 27.08 ± 7.07 13.60 - 39.32 29.39 ± 4.40 19.91 - 37.24 0.272 Knee ROM during Swing (deg) 40.27 ± 9.39 22.83 - 55.14 48.83 ± 5.11 41.10 - 59.15 0.003 Knee Flexion at Initial Contact (deg) 13.47 ± 6.11 5.10 - 22.87 9.59 ± 3.74 5.33 - 20.01 0.037 Knee Flexion at Heel Rise (deg) 10.75 ± 5.71 2.76 - 18.96 11.74 ± 5.36 6.53 - 25.91 0.600 Knee Flexion at Toe Off (deg) 30.94 ± 9.76 13.83 - 48.18 34.72 ± 6.54 21.04 - 45.62 0.201 Knee Peak Flexion (deg) 63.64 ± 10.66 42.52 - 95.71 66.21 ± 6.65 58.36 - 77.88 0.415 Time at Peak Knee Flexion (deg) 71.13 ± 5.52 52.23 - 78.65 73.19 ± 2.17 69.13 - 77.69 0.037 Lateral COM Displacement (cm) 3.71 ± 1.16 1.47 - 5.59 4.35 ± 1.31 3.06 - 7.82 0.127
of each graph. Furthermore, the point for toe off, identifying the division between the stance and swing phase is also visualised on the gait cycle visualisations. The shaded area represents the total observed gait activity while the lighter shaded area represents the gait information of the last measured data. This is to show potential gait improvement in order to assist with rehabilitation evaluation. The example given in figure 4 is the second measurement of subject 2 from Group A. According to the information in the data-base, this subject have suffered an ischemic stroke. Assessment consistent with the clinical data such as decreased knee flexion and abnormality in hip movement is reflected in the normalcy index and the visualisations. Through the visualisation, improvements can also be seen, particularly in regards to knee movement, even though the degree of abnormality is still high.
6
DISCUSSION
6.1
Gait Analysis using the WalkerView
Gait analysis here is done using raw data from the Walkerview 3.0 from TechnoBody, which was originally developed to be used within athletic rehabilitation programs and not aimed for clinical use. As a result, it was unclear to the clinicians how the data provided by the machine can be interpreted for clinical assessment. The aim was thus to develop a processing pipeline to explore the data collected and investigate its potential added value for rehabilitation
assessment. The first step is to figure out how to partition the raw gait data into individual gait cycles.
6.1.1 Automated Gait Segmentation. When analysing gait, first and foremost the data needs to be processed, and one of the key step is to extract gait cycles and identify key gait events. Using an automated method to extract gait cycles is the advantage of digital analysis. Here, the method employed identifies each gait cycle using the periodic hip joint kinematic data through finding the negative peaks of each hip joint in order to obtain the start and end point of a each gait cycle for the opposite leg. The results of the gait cycle extraction is consistent and accurate, indicating a robust algorithm, although there were still potential for errors when dealing with highly irregular or abnormal gait data, and in this case some of that error was corrected manually. Furthermore, since ground reaction force (GRF) data is also available, it was investigated in using GRF data to more accurately partition gait. However, the technically correct way to interpret the GRF data recorded in the raw WalkerView file remained unclear. Through visualising the data and comparing it to available references online, notable patterns were identified. It was determined that data from cell 3 and 4 in particular could be used for identifying the key gait phases loading response and toe off for each gait cycle, as the resulting value obtained through this gait partitioning method yielded results that accurately reflected normative reference data.
Table 4: Incidence of Gait Variable Abnormalities
Gait Variable Decreased (%) Normal (%) Increased (%)
Cadence 6 (28.6) 15 (71.4) 0 (0.0)
Stance Duration (s) 0 (0.0) 13 (61.9) 8 (38.1) Initial Double Support (s) 1 (4.76) 6 (28.6) 15 (71.4) Single Support (s) 1 (4.76) 18 (85.7) 2 (9.52) Terminal Double Support (s) 0 (0.0) 14 (66.7) 7 (33.3)
Stride Time 0 (0.0) 17 (80.1) 4 (19.0)
Trunk Flexion (deg) 0 (0.0) 17 (80.1) 4 (19.0) Trunk Lateral Flexion (deg) 0 (0.0) 16 (76.2) 5 (23.8)
Hip ROM (deg) 4 (19.0) 17 (80.1) 0 (0.0)
Hip Flexion at Initial Contact (deg) 5 (23.8) 16 (76.2) 0 (0.0) Hip Flexion at Heel Rise (deg) 6 (28.6) 13 (61.9) 2 (9.52) Hip Peak Flexion (deg) 0 (0.0) 19 (90.5) 2 (9.52) Knee ROM during Stance (deg) 6 (28.6) 11 (52.4) 4 (19.0) Knee ROM during Swing (deg) 10 (47.6) 11 (52.4) 0 (0.0) Knee Flexion at Initial Contact (deg) 2 (9.52) 14 (66.7) 5 (23.8) Knee Flexion at Heel Rise (deg) 2 (9.52) 18 (85.7) 1 (4.76) Knee Flexion at Toe Off (deg) 7 (33.3) 12 (57.1) 2 (9.52) Knee Peak Flexion (deg) 5 (23.8) 15 (71.4) 1 (4.76) Time at Peak Knee Flexion (deg) 6 (28.6) 13 (61.9) 2 (9.52) Lateral COM Displacement (cm) 1 (4.76) 20 (95.2) 0 (0.0)
Table 5: Normalcy Index Results
Patients Healthy Subject
Normalcy Index Mean Range Mean Range
Gait L 43.375 ± 156.733 0.169 - 720.227 0.933 ± 1.427 0.013 - 5.554 R 20.815 0.278 - 66.428 1.167 0.004 - 4.061 Hip Joint L 89.557 ± 290.436 4.212 - 1327.536 6.533 ± 3.103 2.623 - 12.172 R 67.049 5.985 - 231.266 2.154 2.449 - 10.003 Knee Joint L 126.318 ± 220.339 6.849 - 960.117 7.467 ± 2.376 3.389 - 11.979 R 161.549 7.072 - 743.182 2.648 4.328 - 12.150 Trunk Movement 96.618 ± 115.230 4.401 - 469.462 6.533 ± 1.784 2.753 - 9.380
Thus, the averaged result of the left and right side GRF data and information from load cell 3 and cell 4 in particularly were chosen for the analysis done in this research. Nonetheless, as can be seen in the results in table 1, incidences of irregularity in GRF data is a regular occurrence and was not handled in the proposed gait partitioning algorithm. Potentially in the future, a more robust gait segmentation method that incorporates both kinematic and kinetic data could improve the performance. Furthermore, adding ankle data, which was not collected for this research could potentially help to more accurately determine gait events.
6.1.2 Quantitative Gait Analysis and the Normalcy Index. It is ev-ident that gait is a highly subjective characteristic, which proves a high challenge when evaluating incidences of gait abnormality. Generally, the approach is to compare a patient’s gait to healthy gait samples, which is a similar method the observational method is based on.
It is found through the quantitative analysis of extracted gait variables here that people with acquired brain injuries have multi
joint gait abnormalities [46]. To address the extent of deviation from normality and account for the relationship between correlated gait variables, normalcy indexes (NI) were defined and assessed. The goal of NI is to use one single index to determine how much a subject’s gait deviate from the average normal gait to account for the subjectivity. The measurement is as a result, general in nature, but proves to be a useful tool in objectivity quantifying overall changes in gait, especially in the case for specific joints.
In the case studies performed, the normalcy index proved to be robust enough to clearly indicate incidences of pathological walking patterns, and was able to identify specific leg and joints that suffered injury, and can also assist in the assessment of rehab progress through the assessing the change in the normalcy index. The hip and knee index appears to be the most informative, being able to accurately identify hip and knee pathology as well as subtle performance issues, as is the case with subject 3 of group A where the reported right knee injury is reflected in the normalcy score of the right knee. Furthermore for the case of Group B, the patients have more severe gait dysfunction and had to rely on harnesses to
Figure 4: Gait Evaluation Report Selection: Case Study Group A Subject 2 data 2
perform their walking trial. This may have affected the measure-ment collected, nonetheless the gait analysis also proves to be able to provide accurate assessment, particularly in identifying the main side of the patient where there are issues.
In evaluating extent of improvement, the normalcy index seem to be most useful, as it is shown that NI can help quantify changes in gait in order to evaluate the effects of intervention and treat-ment. However, it should be noted that while NI can help evaluate changes in gait resulting from treatment, it does not necessarily indicate whether there has been any improvement or degradation of function. This is because there is no established relationship between normalcy of gait pattern and function, since abnormalities in gait pattern can represent functional coping mechanisms [17]. Therefore for more complete assessment of the normalcy index naturally will have to be used in conjunction with other outcome measures, and other parts of the neurorehabilitation treatment (see Appendix F).
6.2
Limitations
The sample size available for this study is limited. A key challenge in gait analysis is that there is no standardised gait reference data. Therefore, for each gait analysis research, unique samples need to be collected. As such, due to the small sample size of healthy control subjects with that were collected arbitrarily and do not fit a strict, scientifically controlled demographic, the reliability of the reference has to be doubted. Also, main conventional gait data that would be implemented into quantitative analysis was either not properly utilised (GRF), not present in the raw data (movement speed and distance), not collected (ankle joint data), or not analysed by the WalkerView (kinematic data on the other planes).
7
CONCLUSION
The central question in clinical gait analysis is to understand what a "correct" or "healthy" way of walking is, and to answer the ques-tions: Why do we walk the way we do and why don’t the patients
do so? Has a patient improved? And how so? In a clinical setting, especially one for rehabilitation, the main purpose of gait analysis is to 1) distinguish diagnosis between disease entities, 2) determine severity of disease or injury, 3) assess, evaluate and select treatment options, 4) predict prognosis following intervention or absence of intervention [4]. Quantitative gait analysis can be useful to help identify specific treatment option for specific pathological gait, and make up for the lack of objectivity of observational analyses. Nonetheless, multitudes of challenges are still present, for the lack of standardised normative gait data to review as well as the extent of variability within human gait itself. It should be noted that many disagreements within the medical community still exists when it comes to understanding what affects pathological gait and which abnormal gait characteristics are important to address when decid-ing on intervention. Thus, when approachdecid-ing clinical gait analysis, this should be taken into account.
This exploratory study successfully implemented a gait process-ing pipeline includprocess-ing gait segmentation algorithms and gait feature extraction methods in order to produce clinical analysis profiles for measured subject at the DTC. By comparing gait parameters of the patients to healthy sample subjects, patterns for general abnormali-ties can be observed and evaluated. Additionally, using multivariate statistical methods to summarise 3DGA information into normalcy index proved to be able to provide clinical insight into gait abnor-malities. While limitations such as a lack of coordinated sample data collection strategy and a shortage of reference data is present, the research design and processing pipeline proposed in this research is nonetheless valuable.
This research offers therapists an approach to interpret raw WalkerView data and provides a foundation for future study such as making prediction models for rehabilitation prognosis. The nor-malcy index studied here indicates potential, however it should still be researched and refined in order to determine the most reliable and accurate gait variables for evaluation. Furthermore, a long term data collection strategy should be employed, and should aim to fully utilise all the analytical capacity offered by the WalkerView.
BIBLIOGRAPHY
[1] Mariem Abid, Neila Mezghani, and Amar Mitiche. 2019. Knee Joint Biomechanical Gait Data Classification for Knee Pathology Assessment: A Literature Review. Applied Bionics and Biomechanics 2019 (2019), 1–14. https://doi.org/10.1155/2019/ 7472039
[2] Murad Alaqtash, Thompson Sarkodie-Gyan, Huiying Yu, Olac Fuentes, Richard Brower, and Amr Abdelgawad. 2011. Automatic classification of pathological gait patterns using ground reaction forces and machine learning algorithms. Annual International Conference of the IEEE Engineering in Medicine and Biology Society (2011), 453–457. https://doi.org/10.1109/iembs.2011.6090063
[3] Richard Baker. 2006. Gait analysis methods in rehabilitation. Journal of Neuro-Engineering and Rehabilitation 3, 1 (March 2006), 4. https://doi.org/10.1186/1743-0003-3-4
[4] Richard Baker. 2006. Gait analysis methods in rehabilitation. NeuroEngineering and Rehabilitation 3, 4 (2006).
[5] Matilde Bertoli, Andrea Cereatti, Diana Trojaniello, Laura Avanzino, Elisa Pelosin, Silvia Del Din, Lynn Rochester, Pieter Ginis, Esther M. J. Bekkers, Anat Mirelman, Jeffrey M. Hausdorff, and Ugo Della Croce. 2018. Estimation of spatio-temporal pa-rameters of gait from magneto-inertial measurement units: multicenter validation among Parkinson, mildly cognitively impaired and healthy older adults. Biomed-ical Engineering Online 17, 58 (2018). https://doi.org/10.1186/s12938-018-0488-2 [6] Julien Boudarham, Nicolas Roche, Didier Pradon, Céline Bonnyaud, Djamel Bensmail, and Raphael Zory. 2013. Variations in Kinematics during Clinical Gait Analysis in Stroke Patients. PLoS ONE 8, 6 (June 2013), e66421. https: //doi.org/10.1371/journal.pone.0066421
[7] Julien Boudarham, Nicolas Roche, Didier Pradon, Céline Bonnyaud, Djamel Bensmail, and Raphael Zory. 2013. Variations in Kinematics during Clinical Gait Analysis in Stroke Patients. PLoS One 8, 6 (June 2013), e66421. https: //doi.org/10.1371/journal.pone.0066421
[8] Veronica Cimolin and Manuela Galli. 2014. Summary measures for clinical gait analysis: A literature review. Gait Posture 39 (Feb. 2014), 1005–1010. https: //doi.org/10.1016/j.gaitpost.2014.02.001
[9] Ugo Della Croce, Andrea Cereatti, and Martina Mancini. 2017. Gait Parameters Estimated Using Inertial Measurement Units. Springer, Cham, 1–21.
[10] Jack Crosbie and Roongtiwa Vachalathiti. 1997. Synchrony of pelvic and hip joint motion during walking. Gait & Posture 6, 3 (Dec. 1997), 237–248. https: //doi.org/10.1016/S0966-6362(97)00019-2
[11] Roy B. Davis and Peter A. DeLuca. 1996. Clinical Gait Analysis: Current Methods and Future Directions. Institute of Electrical Electronics Engineers, Chapter 2, 17–38.
[12] A.J. de Kloet, S.R. Hilberink, M.E. Roebroeck, C.E. Catsman-Berrevoets, E. Peeters, S.A.M. Lambregts, F. van Markus-Doornbosch, and T.P.M. Vliet Vlieland. 2013. Youth with acquired brain injury in The Netherlands: a multi-centre study. Brain Injury 27, 7-8 (June 2013), 843–849. https://doi.org/10.3109/02699052.2013.775496 [13] K.J. Deluzio and J.L. Astephen. 2007. Biomechanical features of gait waveform data associated with knee osteoarthritis: An application of principal component analysis. Gait & Posture 25, 1 (2007), 86–93. https://doi.org/10.1016/j.gaitpost. 2006.01.007
[14] Reed Ferber, Sean T. Osis, Jennifer L. Hicks, and Scott L. Delp. 2016. Gait biome-chanics in the era of data science. Journal of Biomebiome-chanics 49, 16 (Dec. 2016), 3759–3761. https://doi.org/10.1016/j.jbiomech.2016.10.033
[15] Joana Figueiredo, Cristina P. Santos, and Juan C. Moreno. 2018. Automatic recognition of gait patterns in human motor disorders using machine learning: A review. Medical Engineering and Physics 53 (2018), 1–12. https://doi.org/10. 1016/j.medengphy.2017.12.006
[16] North American Society for Gait, Human Movement, and AAOP Gait Society. 1993–1994. Terminology of Human Walking.
[17] JR Gage. 1991. Gait analysis in Cerebral Palsy. Clinics in Developmental Medicine 121 (1991).
[18] Salim Ghoussayni, Christopher Stevens, Sally Durham, and David Ewins. 2004. Assessment and validation of a simple automated method for the detection of gait events and intervals. Gait Posture 20, 3 (Dec. 2004), 266–273. https: //doi.org/10.1016/j.gaitpost.2003.10.001
[19] Alessandro Giustini, Caterina Pistarini, and Camilla Pisoni. 2013. Traumatic and nontraumatic brain injury. Handbook of Clinical Neurology 110 (2013), 401–409. https://doi.org/10.1016/B978-0-444-52901-5.00034-4
[20] Gerald F. Harris and Jacqueline J. Wertsch. 1994. Procedures for Gait Analysis. Archives of Physical Medicine and Rehabilitation 75, 2 (Feb. 1994), 216–225. https: //doi.org/10.1002/andp.19053221004
[21] Jeffrey M. Hausdorff. 2007. Gait dynamics, fractals and falls: Finding meaning in the stride-to-stride fluctuations of human walking. Human Movement Science 26, 4 (Sept. 2007), 555–589. https://doi.org/10.1016/j.humov.2007.05.003 [22] Ahmed Refaat Hawas, Heba El-Khobby, Mustafa M. Abd Elnaby, and Fathi E. Abd
El-Samie. 2019. Gait identification by convolutional neural networks and optical flow. Multimedia Tools and Applications 78, 7 (June 2019).
[23] MD Henry G. Chambers and MD David H. Sutherland. 2002. A Practical Guide to Gait Analysis. Journal of the American Academy of Orthopaedic Surgeons 10, 3
(May 2002), 222–231. https://doi.org/10.1080/02699052.2018.1553067 [24] Shuo Jiang, Xingchen Wang, Maria Kyrarini, and Axel Gräser. 2017. A robust
algorithm for gait cycle segmentation. In 2017 25th European Signal Processing Conference (EUSIPCO). IEEE, Kos, Greece. https://doi.org/10.23919/EUSIPCO. 2017.8081163
[25] JA Zeni Jr., JG Richards, and JS Higginson. 2008. Two simple methods for deter-mining gait events during treadmill and overground walking using kinematic data. Gait Posture 27, 4 (May 2008), 710 – 714.
[26] Wasiq Khan and Badii Atta. 2019. Pathological Gait Abnormality Detection and Segmentation by Processing the Hip Joints Motion Data to Support Mobile Gait Rehabilitation. Journal of research in medical sciences 7, 3 (Jan. 2019). https: //doi.org/10.31031/rmes.2019.07.000662
[27] DAvid E. Krebs, Claire E. Robbins, Leroy Lavine, and Robert W. Mann. 1998. Hip Biomechanics During Gait. Journal of Orthopaedic and Sports Physical Therapy 28, 1 (1998).
[28] Sunghyon Kyeong, Seung Min Kim, Suk Jung, and Dae Hyun Kim. 2020. Gait pattern analysis and clinical subgroup identification: a retrospective observa-tional study. Medicine 99, 15 (Feb. 2020), e19555. https://doi.org/0.1097/MD. 0000000000019555
[29] Monika Köhle and Dieter Merkl. 2000. Analyzing human gait patterns for mal-function detection. Proceedings of the 2000 ACM symposium on Applied computing - SAC ’00 1 (March 2000), 41–45. https://doi.org/10.1145/335603.335629 [30] Jennifer L McGinley, Patricia A Goldie, Kenneth M Greenwood, and Sandra J
Olney. 2003. Accuracy and Reliability of Observational Gait Analysis Data: Judgments of Push-off in Gait After Stroke. Physical Therapy 83, 2 (Feb. 2003), 146–160. https://doi.org/10.1093/ptj/83.2.146
[31] John W. Michael and John H. Bowker. 1992. Normal Gait. American Academy of Orthopedic Surgeons, Chapter 13.
[32] Brain Injury Association of America. 2020. ABI vs. TBI: What is the difference? https://www.biausa.org/brain-injury/about-brain-injury/nbiic/what-is-the-difference-between-an-acquired-brain-injury-and-a-traumatic-brain-injury [33] Brain Injury Association of Canada. 2007. Acquired Brain Injury Information.
Re-trieved September 25, 2016 from www.braininjurycanada.ca/wp-content/uploads/ 2007/05/BIAC-Fact-Sheet-2014.pdf
[34] Jacqueline Perry and Judith M. Burnfield. 2010. Gait Analysis: Normal and Pathological Function. Journal of Sports Science Medicine 9, 2 (June 2010), 3–47. [35] Cauchy Pradhan, Max Wuehr, Farhoud Akrami, Maximilian Neuhaeusser,
Sab-rina Huth, Thomas Brandt, Klaus Jahn, and Roman Schniepp. 2015. Automated classification of neurological disorders of gait using spatio-temporal gait pa-rameters. Journal of Electromyography and Kinesiology 25 (2015), 413–422. https://doi.org/10.1016/j.jelekin.2015.01.004
[36] Scientific research project proposal. [n.d.]. Quality Care Program for Intensive Neurorehabilitation after Severe Brain Injury in Young Adults - Development of pre-processing and analysis pipelines for structured clinical data to facilitate predictive modelling.
[37] Jan Rueterbories, Erika G Spaich, Birgit Larsen, and Ole K Andersen. 2010. Meth-ods for gait event detection and analysis in ambulatory systems. Medical Engi-neering Physics 32, 6 (July 2010), 545–552. https://doi.org/10.1016/j.medengphy. 2010.03.007
[38] L.M. Schutte, U. Narayanan, J.L. Stout, P. Selber, J.R. Gage, and M.H. Schwartz. 2000. An index for quantifying deviations from normal gait. Gait and Posture 11, 1 (Feb. 2000), 25–31. https://doi.org/10.1016/S0966-6362(99)00047-8 [39] Ki Young Shin, Yonghoon Rim, Youn Soo Kim, Hyo Shin Kim, Jae Woong Han,
Chang Hyun Choi, Kyung Suk Lee, and Joung Hwan Mun. 2010. A Joint Normalcy Index to Evaluate Patients with Gait Pathologies in the Functional Aspects of Joint Mobility. Journal of Mechanical Science and Technology 24, 9 (Oct. 2010), 1901–1909. https://doi.org/10.1007/s12206-010-0608-9
[40] Robert W. Soutas-Little. 1998. Gait Analysis and Biomechanics. Chapter 2, 49–68. [41] SpinalCord.com. 2019. Acquired Brain Injuries. https://www.spinalcord.com/
acquired-brain-injury
[42] TechnoBody 2019. Walker View 3.0 SCX | TechnoBody. https://www.tecnobody. com/en/products/detail/walker-view-3-scx
[43] Manan Vora, Digpal Ranawat, Manit Arora, and Apoorva Tiwari. 2019. Run and Gait Analysis in Junior Badminton Players of South Asia. Journal of Physical Fitness, Medicine Treatment in Sports 6, 4 (July 2019). https://doi.org/0.19080/ JPFMTS.2019.06.555695
[44] Gavin Williams, Brook Galna, Meg E. Morris, and John Olver. 2010. Spatiotem-poral Deficits and Kinematic Classification of Gait Following a Traumatic Brain Injury: A Systematic Review. The Journal of Head Trauma Rehabilitation 25, 5 (Sept. 2010), 366–374. https://doi.org/10.1097/htr.0b013e3181cd3600 [45] Gavin Williams, Daniel Lai, Anthony Schache, and Meg E Morris. 2015.
Classi-fication of gait disorders following traumatic brain injury. The Journal of head trauma rehabilitation 30, 2 (April 2015), e13–23. https://doi.org/10.1097/HTR. 0000000000000038
[46] Gavin Williams, Meg E. Morris, Anthony Schache, and Paul R. McCrory. 2004. Incidence of Gait Abnormalities After Traumatic Brain Injury. Archives of Physical Medicine and Rehabilitation 90, 4 (April 2004), 587–593. https://doi.org/10.1016/j. apmr.2008.10.013
[47] Terri Wilson, Olivia Martins, Michelle Efrosman, Victoria DiSabatino, B Mohamed Benbrahim, and Kara K Patterson. 2018. Physiotherapy practice patterns in gait rehabilitation for adults with acquired brain injury. Brain Injury 33, 3 (Dec. 2018), 333–348. https://doi.org/10.1080/02699052.2018.1553067
[48] Sławomir Winiarski and Alicja Rutkowska-Kucharska. 2009. Estimated ground reaction force in normal and pathological gait. Acta of bioengineering and biome-chanics 11, 1 (2009), 53–60.
[49] Jianning Wu, Jue Wang, and Li Liu. 2007. Feature extraction via KPCA for classification of gait patterns. Human Movement Science 26, 3 (2007), 393–411. https://doi.org/10.1016/j.humov.2007.01.015
ACKNOWLEDGEMENT
I would like to give special thanks to Ruud van der Veen and dr. Erik Bekkers for providing primary supervision during this project period and accommodated me throughout the process. I would also like to thank UvA, UMC and the Daan Theeuwes Centrum along with dr. Frank Nack, dr. Marsh Königs, study advisors Cecilia Sigvardsdotter, Sophie Tjebbes and fellow student Wietske Dotinga, Sean Hladkyj for their support during the project. Lastly I extend my gratitude to everyone who believed in me and provided me with assurance throughout, this has been quite a year, quite a year indeed.
APPENDIX
A
GAIT ANALYSIS REFERENCE
Figure 5: Normative Gait Data
Figure 6: Gait Cycle and Critical Gait Events Summary
B
WALKERVIEW DATA REPORTING
Figure 7: walkerview raw data top to bottom: (1,2) GRF, (3) Kinematics, (4,5) Hips, (6,7) Knees, (8,9) Trunk, (10) Load, (11) COG
Figure 8: WalkerView Gait Analysis Report Overview Front Page
Figure 9: WalkerView Gait Analysis Report Overview Gait Cycle Visualisation
C
GAIT SEGMENTATION
Figure 10: Gait Segmentation Results Visualised
D
GAIT VARIABLES
Table 6: Gait Variables and Descriptors
Parameters Unit Description SpatioTemporal
Step Time sec average step duration Stride Time sec average stride duration
Cadence steps/min average steps taken per minute
Stance Phase duration sec & (%GC) percentage of stance phase per each gait cycle Swing Phase duration sec & (%GC) percentage of swing phase per each gait cycle
Single Support sec & (%GC) Time duration in GC when single foot is bearing body weight Double Support sec & (%GC) Time duration in GC when both feet and bearing body weight Initial Double Support sec duration of initial double support phase
Terminal Double Support sec duration of terminal double support phase Kinematic
Trunk Flexion deg range between maximum and minimum value in the gait cycle Trunk Lateral Flexion deg range between maximum and minimum value in the gait cycle Hip & Knee ROM deg range of motion of hip and knee joints during each gait phase
Hip Extension deg peak value at terminal stance Knee Flexion IC deg angle at initial contact Knee Flexion MS deg angle at mid stance Knee Flexion Swing deg angle at toe off
Vertical Displacement COG cm range between maximum and minimum in the gait cycle
E
NORMALCY INDEX
Table 7: Gait Variables for normalcy index
Gait Parameter Gait Phase Unit Correlation 1 Stance Time Stance %GC 0.576 2 Cadence Stance/Swing step/min 0.497 3 Hip Peak Extension Stance/Swing deg 0.046 4 Hip Peak Flexion Swing deg 0.251 5 Hip ROM Stance/Swing deg 0.938 6 Knee Flexion Initial Contact deg 0.410 7 Knee Flexion Toe Off deg 0.917 8 Time Peak Knee Flexion Swing %GC -0.176 9 Knee ROM Stance/Swing deg -0.028 10 COM Displacement Stance/Swing cm -0.209
Table 8: Hip variables for normalcy index
Gait Parameter Gait Phase Unit Correlation 1 Hip Peak Extension Stance/Swing deg 0.880 2 Hip Peak Flexion Swing deg 0.832 3 Hip ROM Stance deg 0.948 4 Hip ROM Swing deg 0.905 5 Hip std ROM Stance/Swing deg 0.886 6 Hip std Minima Stance/Swing deg 0.669 7 Hip std Maxima Stance/Swing deg 0.931 8 Hip Flexion Initial Contact deg 0.612
Table 9: Knee Variables for normalcy index
Gait Parameter Gait Phase Unit Correlation 1 Knee Peak Flexion Swing deg 0.844
2 Knee ROM Swing deg 0.845
3 Knee std ROM Stance/Swing deg 0.954 4 Knee std Minima Stance/Swing deg 0.931 5 Knee std Maxima Stance/Swing deg 0.949 6 Knee Flexion Initial Contact deg 0.934 7 Knee Peak Flexion Time Swing deg 0.865 8 Mean Flexion Extension Velocity Loading Response deg/s 0.962 9 Minimum Flexion Extension Velocity Terminal Stance deg/s 0.942
Table 10: Trunk Variables for normalcy index
Gait Parameter Gait Phase Unit Correlation 1 Trunk ROM Swing deg 0.202 2 Trunk ROM Stance deg 0.254 3 Trunk ROM Stance/Swing deg 0.315 4 Trunk std Minima Stance/Swing deg 0.177 5 Trunk std Maxima Stance/Swing deg 0.067 6 Trunk Lateral ROM Swing deg 0.445 7 Trunk Lateral ROM Stance deg 0.369 8 Trunk Lateral ROM Stance/Swing deg 0.423
Figure 11: Normalcy Index Scores
F
MEASUREMENTS TAKEN IN NEUROREHABILITATION PROGRAM AT THE DTC
Figure 12: Physical, Occupational, Speech, Neuropsychological Therapy Measurements
Figure 13: Medical Rehabilitation and Social Rehabilitation Measurements
