Performance-based selective training for robot-mediated upper limb motor learning and stroke rehabilitation

Stroke is a major cause of disability in the adult population, costing the European economy over Euro30 billion per year. The majority of stroke survivors suffer from sensory and movement deficits that result in long-term, and sometimes severe, movement disorders. Partial paralysis or weakness of the upper limb with impaired execution of movements is common, leaving many stroke patients unable to easily perform everyday actions such as reaching and grasping for objects, and so limiting their quality of life.

A recent meta-analysis summarising several independent studies has concluded that given the balance of benefits and risks, robot-assisted rehabilitation should be provided to US outpatients and chronic stroke survivors, with a "robot" guiding their movements and providing a regulated practice regime. These robotic devices are motorized, jointed, levers, the handle of which is held by the patient or attached to the arm, and the motors provide gentle forces that guide the desired movement, Although beneficial, the practice schedules used in these studies are typically quite simple and rarely take account of the patient's individual abilities, unlike standard, practitioner-led rehabilitation. In contrast, a recent USA study has shown that individually adjusted performance-based practice within a robot-assisted regime can be beneficial, facilitating and retaining recovery of the affected arm's movement. This practice regime, where the arm movement was initially guided by the robot, and the amount of assistance was reduced over time, led to substantially greater motor improvement for a given period of rehabilitation compared to other robotic therapies. But the SELECTION of movements that were trained was still regulated, and did not reflect the patient's abilities. It is already known that practice of a task that is varied is more effective for training than practice of a consistent task. Moreover, if individual impairment/performance is not taken into account in the selection of practiced movement, there is a risk that practice might lead to worsened motor function (e.g. by encouraging pathologic movement patterns).

We therefore aim to test the advantages of combining individually tailored selection of practice movements, to take account of the patients' unique and changing levels of performance, with state-of-art robot-assisted therapy. To do so, we will first expand on our existing pilot data showing that an algorithm that selects movement practice conditions based on individual performance "maps" can generate the most learning. This will first be tested using unimpaired, healthy participants across a number of different practice regimes to confirm the most advantageous training. We will also test that this learning generalises to unpracticed movements, and is retained for at least a week after practice.

We will next study the relationship between these 2D movement "maps" and the underlying upper arm joint movements and major muscle activity, to better understand the relationship between the limb's biomechanics, the neural control of muscles, and the performance maps. We will study these relationships in both healthy participants, to build up a template of normal performance, and in stroke survivors.

Finally we will test the benefit of our adaptive, individuated, and performance-based training on stroke survivors, testing for sustained improvement over 16 weeks of training, testing for generalisation of the improved performance to unpracticed movements, and quantifying the improved movement and its transfer from the lab into daily actions using established clinical scoring schemes. The results of our project would then provide the basis for a full clinical trial, using larger groups of patients, as a follow up study.

Stroke is a leading cause of disability, with rising incidence as the western population ages. Stroke morbidity costs the EU economy more than an estimated Euro30b p.a. Over half of stroke survivors suffer from sensory and motor deficits. Upper limb hemiparesis is common, leaving the patients chronically - and sometimes severely - impaired in planning and/or executing movements with their paretic hand (e.g. reaching and grasping), reducing independence and quality of life.

Recent research suggests that robot-assisted rehabilitative practice improves upper limb function, but current regimes focus almost exclusively on fixed movement paradigms. It is likely that learning and recovery also depend on the selection of practiced movements, varied across individuals, and reflecting their individual abilities. Since current robot-assisted therapy does not involve such movement selection, therapist-assisted rehabilitation is both more patient focussed and more interactive.

We suggest, based on pilot data, that individual tailoring of the selection of robot-assisted movements based on patient performance will benefit recovery, and that motor recovery will be most effective when the practice movements are selected from regions of the individual's performance "map" where the change between best and worst performance is most steep (spatially), allowing generalization of training from the best areas into the worst areas.

We aim to test (1) that most learning is induced when practice is focused around the steepest local performance gradients; (2) any additional improvement by changing task difficulty across sequential training sessions. We will characterise (3) correlations between the pattern of reaching performance in task space i.e. to test the reliability of the performance map as an index of internal postural parameters; (4) learning-related changes in the parameter space. We will (5) quantify the benefit of performance-based practice to hemiparetic stroke patients.

The work proposed here is essentially basic science, addressing important questions about how performance-based practice can affect human motor learning and the recovery of motor abilities after stroke. So in the short term, its impact will be principally on researchers in the fields of human motor control and learning theory and of robot-assisted rehabilitation. However, the research is designed to lead onto a larger scale clinical assessment of our robot-assisted practice regimes. Eventually, by furthering our understanding of these practice regimes, we expect to impact on other areas of neurology and rehabilitation. This is expected to have greater long term impact on rehabilitation procedures and on stroke patient recovery. The beneficiaries will therefore be those health-related professionals employed for example in the NHS involved in assisting patients with impaired limb function, whether those patients are otherwise healthy or are suffering a more generalised deterioration in health such as the elderly. It is possible that a reduction in cost would accrue from these improved rehabilitation regimes through reduction in time spent by the professionals with each patient and through a reduction in the long term assistance required by these patients, due to better-optimized patient-tailored practiced.

The training and tasks we will use in the project are necessarily focused around upper limb movement, and visually guided reaching actions. However, the motor learning mechanisms by which such recovery of upper limb function is achieved are likely to be applicable to other systems, including lower limb function and gait. Furthermore, the principle that taking account of relations between performance maps and training selection is likely to be relevant to, and probably common to, other types of motor learning. Thus, the research may be of a general relevance to sport training and sport rehabilitation, possibly leading to increased duration in the careers of sportsmen and women as well as maintaining the capacity of the general population to enjoy sport for a greater part of their lives, and to training to preserve healthy function in old age.

Manufacturers of robotic equipment will benefit from this research, validating their equipment for use in these applications and providing opportunities for further development to satisfy improvements achievable as identified by the research in this project.