Towards personalized robot-assisted gait training
Due to the COVID-19 crisis measures the PhD defence of Simone Fricke will take place online without the presence of an audience.
The PhD defence can be followed by a live stream.
Simone Fricke is a PhD student in the research group Biomedical Engineering. Her supervisor is prof.dr.ir. H. van der Kooij from the Faculty of Engineering Technology.
Robot-assisted gait training (RAGT) is a promising tool to improve walking function after stroke and spinal cord injury (SCI), especially when combined with conventional physical therapy. The way how the robot is controlled can have a large influence on active participation of the user and the effectiveness of the training. Previous studies suggest that personalized assistance based on patients' abilities can be beneficial as it can increase active participation of the user. Besides, robotic gait trainers cannot only provide training, but also have the potential to be used for assessment of walking function and impairments affecting walking function. The goal of this thesis is to take a next step towards personalized robot-assisted gait training by 1. developing assessment tools to quantify walking function and underlying impairments affecting walking function, and 2. improving subtask-based assistance and optimizing assistance tuning based on users' walking abilities.
Knowledge about the mechanical properties of the joints (i.e. joint impedance) can be used to monitor patients' progress, improve training protocols and adjust controllers for RAGT. While ankle joint impedance during walking has been determined in previous studies, hip joint impedance during walking is unknown. The goal of Chapter 2 is to develop and evaluate a device that can be used to determine hip joint impedance during walking, and to get first estimates of apparent hip joint impedance during walking. We developed the LOPER (LOwer limb PERturbator) which has negligible effects on the walking pattern when attached to the upper leg, and can apply force perturbations during swing phase. Through application of perturbations at different instances of the swing phase, we were able to get first estimates of the time-varying behaviour of the apparent hip joint impedance during walking in healthy participants. In the future, the device and applied estimation methods should be evaluated in people with neurological disorders. In addition, the device and estimation methods should be extended to the knee joint to be able to estimate knee joint impedance and possible interactions between knee and hip joint impedance.
The minimal assistance that a user needs to walk in a robotic gait trainer can be a measure of the user's walking function. The lower this assistance is, the better the walking function. Controllers that automatically adjust assistance based on users' performance could be used as assessment tools for walking function. However, results from current automatically-tuned (AT) algorithms can be difficult to interpret, as assistance is adjusted for several intervals of the gait cycle which are not directly related to functional aspects of gait. In addition, it is not known how the applied assistance is affected by changes in training parameters, such as partial body weight support (PBWS) and walking speed, that can occur during therapy. In Chapter 3, we describe a controller that automatically adjusts the assistance for various subtasks of walking (e.g. foot clearance, stability during stance) based on users' performance. The effects of changes in PBWS and walking speed on the applied assistance and subtask performance were evaluated. Experiments in ten healthy participants showed that subtask performance, and thus applied assistance, was influenced by both, PBWS and walking speed. The applied assistance by the AT subtask-based controller can be used as an assessment tool of walking function, but only if PBWS and walking speed are kept constant.
Manual tuning of subtask-based assistance can be difficult, time-intensive and depends on subjective decisions of therapists. Automatic assistance tuning can tackle these problems, but potentially has drawbacks, too. The goal of Chapter 4 is to determine differences and similarities between manually- and automatically-tuned (MT and AT) subtask-based assistance in people with neurological disorders (stroke or SCI). Participants' preferences, time to tune the assistance, final assistance levels and errors compared to reference trajectories were analyzed. Participants did not prefer one approach (AT or MT) over the other regarding comfort, safety, and amount and effect of applied assistance. We found several advantages of AT assistance compared to MT assistance in people with neurological disorders who were walking in the LOPES II gait trainer: quicker assistance tuning, lower assistance levels and separate tuning of each subtask resulting in a good performance for all subtasks. Clinical trials are needed to show whether these apparent advantages of AT assistance also result in better clinical outcomes.
In Chapters 3 and 4, automatic assistance tuning was performed for all subtasks simultaneously. We did not consider whether assisting one subtask (e.g. the most impaired one) would also affect other subtasks. In Chapter 5, we aimed to get a better understanding of separate assistance for the most impaired subtasks of walking after stroke: foot clearance (FC), stability during stance (SS) and weight shift (WS). Performance for the impaired, assisted subtasks clearly improved in mildly impaired chronic stroke survivors compared to walking in LOPES II without assistance. Our WS assistance can be further optimized so that users shift the weight better towards the paretic leg. Performance improvements for the assisted subtasks only rarely resulted in more general changes of the walking pattern, i.e. effects on other subtasks or spatiotemporal parameters. Therefore, in mildly impaired stroke survivors, assistance for various subtasks of walking can be tuned simultaneously resulting in quick assistance tuning. There is no need for specific, time-intensive, tuning protocols in these patients, such as tuning subtasks after each other while starting with the most impaired subtask.
To sum up, we developed assessment methods and a control algorithm that automatically adjusts subtasks-based assistance based on users' performance. We have taken a next step towards personalized robot-assisted gait training. In the discussion we show that there are still some challenges to overcome in order to apply the optimal robotic gait assistance for each individual. Future research should focus on the long-term effect of various controllers and a better understanding of the exact effect of RAGT on neurorehabilitation after stroke and SCI to further personalize and improve RAGT.