Matthew Mavor
Research
Figure 1. Calculation process for determining the maximum finite-time Lyapunov exponent from wearable IMU sensors and optical motion capture data (Beange et al., 2019).
Figure 2. Motion capture photo progression of patient with adolescent idiopathic scoliosis performing a forward bending movement while being tracked with a custom optical intersegmental spine model (Maksimovic et al., 2021)
Figure 5. Overview of the different movement task and the IMU placement used to measure spine control in patients with low back pain (A). The flexion/extension task in the sagittal plane; (B) the rotational task in the transverse plane;(C) the complex task in three dimensions; and (D) the placement of the IMU's (Graham et al., 2020).
Figure 1. Calculation process for determining the maximum finite-time Lyapunov exponent from wearable IMU sensors and optical motion capture data (Beange et al., 2019).
Spine Biomechanics and Neuromuscular Control
Our lab assesses how spine neuromuscular control and stability interact with biomechanical (e.g. spine loading, fatigue), physiological (e.g. inflammatory blood markers), and psychological (e.g. kinesiophobia) factors to influence the risk of developing low back disorders during spine movement tasks (e.g. repetitive lifting). Moreover, we study spine neuromuscular control impairments and movement strategies in both induced and chronic low back pain (LBP), as well as changes in control that can occur as the result of interventions (e.g. strength or muscle activation training) and treatments (e.g. physiotherapy and surgery). Finally, as a lab we have been early adopters of dynamical systems analyses and techniques (e.g. local dynamic stability, continuous relative phase, etc.) to study spine neuromuscular control and stability during dynamic movement tasks. However, since there are multiple techniques used within the biomechanics and rehabilitation fields for the quantification of spine neuromuscular control and stability, we are proactive in developing collaborative relationships to compare/contrast our techniques with other methodologies (e.g. systems identification) to ensure our work is continuously evolving, valid, and at the cutting-edge of the field.
For more information check out the biographies of Tianna, Kristen and Victor
Novel Measurement Tools and Techniques
Many techniques used in biomechanical research require expensive and immovable equipment, which reduces the utility of this equipment to most clinicians, ergonomists, and coaches. Therefore, an aim of the lab is to develop novel measurement tools and techniques that are able to be implemented in the field for clinical, ergonomic, research, and sport use. In order for widespread use, the novel measurement tools and techniques need to be affordable, transportable, flexible, reliable, and easily interpretable. As a lab, we are continuing to develop IOS and android applications that combine the use of wearable sensors (inertial measurement units; IMUs) with cloud-based data checking and computations to assess spinal control/movement quality measures to assist in the diagnosis and treatment of low back disorders in the clinic. We are also integrating the use of video cameras and depth sensors paired with convolutional neural networks as alternatives to expensive motion capture systems for field-based movement assessments. In addition, the trunk is often represented as either a one- or two-segment rigid body; however, each intervertebral joint is capable of its own independent motion. Therefore, we have been working with a novel intervertebral model that is able to analyze motion at each intervertebral joint for a deeper kinematic understanding of differences between low back disorders and to assist in the diagnosis of low back disorders in the clinic.
For more information check out the biographies of Gwyneth, Jessica and Kristen.
Figure 1. Screenshots displaying the workflow of a simple working version of our laboratory app that collects IMU data and shares it via a number of options (Beange et al., 2019)
Figure 2. Framework for subgrouping low back pain patients using wearable motion sensors, cloud-based computing and big data analytics (Beange et al., 2017).
Figure 5. Spatial Reconstruction of the intervertebral kinematic model depicting an A) sagittal, B) posterolateral and C) posterior viewpoint from a representative participant (Beaudette et al., 2019, 2020).
Figure 1. Screenshots displaying the workflow of a simple working version of our laboratory app that collects IMU data and shares it via a number of options (Beange et al., 2019)
Figure 1. Overarching lab framework for assessing movement quality in clinical, ergonomic, and sport settings (Ross et al., 2018).
Figure 2. Reconstruction of a linear discriminant function differentiating elite and novice athletes during the T-balance right movement at 0%, 50%, and 100% of the movement (Ross et al., 2018).
Figure 5. Inertial motion capture of military personnel performing functional activities under various body-borne loads (Mavor et al., 2020).
Figure 1. Overarching lab framework for assessing movement quality in clinical, ergonomic, and sport settings (Ross et al., 2018).
Movement Strategy and Quality Assessment
Movement screens are used to identify abnormal movement patterns that are 1) indicative of dysfunction, or 2) may increase the risk of injury/hinder performance. Abnormal patterns are traditionally detected through visual observations by a coach, clinician, or ergonomist; however, there are limitations to these subjective movement screens (i.e. poor inter and intra-rater reliability, differences need to be large enough to be seen by the observer, etc.). Therefore, we are developing and refining quantitative data-driven methods to reduce issues related to reliability, and offer the potential to detect new and important features that may not be observable by the human eye. We are currently applying pattern recognition techniques to whole-body kinetic and kinematic data (e.g. principal component analysis) to detect important features that can then be used for more advanced analyses. Currently, we are developing new techniques and methods for detecting movement strategies and quality in athletes, employees (e.g. military), and patients (e.g. low back pain patients).
For more information on current research on this topic, check out the biographies of Gwyneth, Matthew and Xiong.
Neuro-Musculoskeletal and Digital Human Modelling
Advancements in computing technology have brought in silico simulations to the leading edge of human movement analysis. Neuro-musculoskeletal modelling allows researchers and clinicians to assess muscle force generation and joint loading in an effort to delineate cause-and-effect relationships. One commonly used platform is an open-source software called OpenSim. We have been working to improve OpenSim models for assessing spine biomechanics and ergonomic tasks such as lifting. Recent work has been aimed at: 1) developing and validating an improved full-body lumbar spine model for the assessment of spine loading during lifting tasks, 2) developing an improved MATLAB/OpenSim API for incorporating stability constraints within any OpenSim musculoskeletal model, 3) developing and validating new methods for modeling the interaction between the body and external loads to improve simulation accuracy, and 4) working with OpenSense to accurately drive a model of gait variability using inexpensive IMU sensors. In silico simulations are also important for digital human modelling. We have created a data driven framework that incrementally morphs collected movement patterns in our military to iteratively represent a wide variety of personal and body-borne load characteristics and then recreate them in virtual environments. This method provides an opportunity to accurately model movements within a wide range of virtual environments such as battle simulators, virtual reality, custom software, and musculoskeletal modelling platforms.
For more information on current research in this topic, check out the biographies of Mohammad, Matthew and Isabel and Chris
Figure 1. Neuro-musculoskeletal modelling of a participant performing different lifting activities. Both the participant's and the external load's motion were tracked using optical motion capture (Akhavanfar et al., 2021a).
Figure 2. Schematic of approaches to modelling external loads in OpenSim for two-handed lifting. Blue and red borders indicate the methods used for inverse kinematics (IK) and static optimization (SO) calculations, respectively. Green denotes OpenSim bodes (Akhavanfar et al., 2021b).
Figure 5. Military movement patterns are morphed to incrementally represent body-borne load conditions (~37, ~23, ~5.5 kg presented) and comprehensively analyzed in OpenSim. L4/L5 AP Shear (left) and Compression (right) graphically displayed (Mavor et al., 2021).
Figure 1. Neuro-musculoskeletal modelling of a participant performing different lifting activities. Both the participant's and the external load's motion were tracked using optical motion capture (Akhavanfar et al., 2021a).
Pattern Recognition and Machine Learning
As data sets continue to grow in size, we have been implementing pattern recognition techniques (e.g. principal component analysis) and machine learning to identify important features within a data set and then classify these data based on different classifiers (e.g. risk of injury, movement type, skill level). For example, we are using these methods to classify data within ergonomic (i.e. military), clinical (e.g. low back pain assessment), and sport (e.g. screening athletes) settings to help with injury prediction, identification, and rehabilitation. We have also used machine learning to develop activity classification algorithms, novel markerless tracking methods, as well as new software for the automatic labelling of motion capture markers. We use many different machine learning techniques such as binary logistic regression, linear discriminant analyses, support vector machines, k-nearest neighbors, clustering, LSTM, convolutional neural networks, and more.
For more information on current research in this topic, check out the biographies of Victor, Xiong, Gwyneth, Matthew, Kristen, Allison.
Figure 1. An open-source software package was created that automatically labels optical motion capture markers using deep learning (Clouthier et al., 2021a).
Figure 2. Differences in lumbar spine geometry between those who are schedules to undergo spinal fusion surgery and those who are not can be identified using statistical shape modelling (Clouthier et al., 2021b).
Figure 5. The percent of correctly classified athletes as either a elite or novice for when 1 to the total number of PCs retained were retained for binary logistic regression (BLR), decision tree (DT), linear discriminant analysis (LDA), k-nearest neighbours (kNN), naive Baynes (NB), support vector machine with a linear kernel (SVM), and support vector machine with radial basis function kernel (RBF) and leave-one-out validation for simulated inertial measurement unit data (Ross et al., 2020)
Figure 1. An open-source software package was created that automatically labels optical motion capture markers using deep learning (Clouthier et al., 2021a).