Final Report Summary - MARS (Modeling Arm Recovery after Stroke)
The two scientific objectives of this IIF proposal were to: 1. Develop and validate a multiple-time scale computational model of recovery of arm function after stroke based on a scientific understanding of neural plasticity and motor learning after brain injury. 2. Develop a novel adaptive and personalized motor trainer to improve arm function after stroke based on predictions of long-term recovery from the model.
Results Objective 1: We have collected data of 30 healthy subjects and 12 subjects post-stroke to study motor learning and recovery at both hand- and arm joint-level. Post-stroke subjects typically develop abnormal couplings, or “synergies”, between joints post-stroke. Studying these synergies is important, because it is conceivable that subjects post-stroke recover very well in the task/hand space, generating healthy-like movements, while still using compensatory “abnormal” movements in the joint level. We have devised a new method to analyze these synergies; previous methods are based on principal component analysis of joint angles – these methods do not take into account the movement of the hand into account – thus, because the arm is redundant, there could be movements in some joints that do not affect hand movements. Our new method can parse out the hand-related joint movements from those joint movements that have no effect on the hand. Our method therefore allows us to model recovery post-stroke at both the joint and the hand levels. We then found that motor training in the sub-acute phase post-stroke led to improvements in hand-space that are characterized by three different time constants: within-session, across days, and across weeks. In addition, we found a relationship between within-session improvement and long-term recovery.
Results Objective 2: We have devised a new scheduling method that maximizes predicted long-term gains by optimizing the presentation of multiple motor tasks in healthy subjects. We have used optimal control theory to select the schedules that maximize long-term retention based on computational models of motor adaptation. Previous models show that motor adaptation is due to a combination of fast and slow processes, with long-term memory being the result of activity in the slow process. We therefore modeled adaptation of different tasks with such slow/fast model to predict long-term retention; the optimal schedule then determined the sequence of tasks presentation that maximized long-term retention for all tasks. We have tested this scheduling method in a number of training duration, task difficulty and interfering conditions.
Conclusions and socio-economic impacts. With regards to Objective 1, the next steps are to validate individual predictive models of recovery with multiple time scales. With regards to Objective 2, the next steps are to use these validated models to develop optimal schedules of motor training that maximize long-term recovery post-stroke. The impact of this research will be important for patients, clinicians, and insurance companies, because such models and scheduling methods will allow us to determine the timing and dose of motor training post-stroke.
Results Objective 1: We have collected data of 30 healthy subjects and 12 subjects post-stroke to study motor learning and recovery at both hand- and arm joint-level. Post-stroke subjects typically develop abnormal couplings, or “synergies”, between joints post-stroke. Studying these synergies is important, because it is conceivable that subjects post-stroke recover very well in the task/hand space, generating healthy-like movements, while still using compensatory “abnormal” movements in the joint level. We have devised a new method to analyze these synergies; previous methods are based on principal component analysis of joint angles – these methods do not take into account the movement of the hand into account – thus, because the arm is redundant, there could be movements in some joints that do not affect hand movements. Our new method can parse out the hand-related joint movements from those joint movements that have no effect on the hand. Our method therefore allows us to model recovery post-stroke at both the joint and the hand levels. We then found that motor training in the sub-acute phase post-stroke led to improvements in hand-space that are characterized by three different time constants: within-session, across days, and across weeks. In addition, we found a relationship between within-session improvement and long-term recovery.
Results Objective 2: We have devised a new scheduling method that maximizes predicted long-term gains by optimizing the presentation of multiple motor tasks in healthy subjects. We have used optimal control theory to select the schedules that maximize long-term retention based on computational models of motor adaptation. Previous models show that motor adaptation is due to a combination of fast and slow processes, with long-term memory being the result of activity in the slow process. We therefore modeled adaptation of different tasks with such slow/fast model to predict long-term retention; the optimal schedule then determined the sequence of tasks presentation that maximized long-term retention for all tasks. We have tested this scheduling method in a number of training duration, task difficulty and interfering conditions.
Conclusions and socio-economic impacts. With regards to Objective 1, the next steps are to validate individual predictive models of recovery with multiple time scales. With regards to Objective 2, the next steps are to use these validated models to develop optimal schedules of motor training that maximize long-term recovery post-stroke. The impact of this research will be important for patients, clinicians, and insurance companies, because such models and scheduling methods will allow us to determine the timing and dose of motor training post-stroke.