Copyright �� Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. memorized and then replayed. It is the goal not the movement that has to be repeated. When we reach for a glass of water we do so differently each time because of small differences in posture, position of the glass relative to the body etc. Nevertheless, a reach is always success- fully achieved. Within the context of the practice schedule, it may be that the influential idea of a motor recovery plateau at 6 months after stroke reflects asymp- totic learning after massed practice rather than a true biological limit [14 ]. Motor learning in patients with hemiparesis There have been surprisingly few studies of motor learn- ing after stroke and almost none looking at deficits in motor memory formation despite the likely relevance of these processes to rehabilitation [15] and recovery [16]. Winstein and colleagues [17] tested the ipsilesional arm in patients with middle cerebral artery territory infarctions using an extension-flexion elbow reversal task on a horizontal surface with feedback given as knowledge-of-results. The authors found no difference in acquisition on day 1 or recall on day 2 between patients and controls although patients were less accu- rate overall. Only the ipsilesional arm was tested, how- ever, so a learning deficit in the affected arm was not evaluated. A more recent study [18] showed impaired adaptation of the paretic arm to a laterally displacing force-field generated by a robot arm. Patients showed reduced capacity to make straight movements in the force-field and showed reduced after-effects. The authors concluded, however, that patients did not have a learn- ing deficit per se but weakness-related slowness to develop the required force to implement anticipatory control. Thus, at the current time it remains uncertain whether there are specific motor learning deficits in patients with hemiparesis. There are a number of reasons for this. First, there have been too few studies. Second, there are many types of motor learning and they may be differentially affected depending on lesion location. Third, it can be difficult to demonstrate a learning abnormality in patients when performance is already considerably impaired at baseline. Is recovery from hemiparesis a form of motor learning? Longitudinal studies suggest that recovery from hemiparesis proceeds through a series of fairly stereo- typical stages over the first 6 months post-stroke, irre- spective of the kind of therapeutic intervention [19]. In particular, although there is heterogeneity in stroke severity and recovery across individuals, it has been shown that the time course of the change in the Barthel Index for patients with middle cerebral artery stroke is well fitted by a logistic regression model [20]. The model indicates that the earlier that patients show recov- ery, the better the outcome at 6 months and that the Barthel Index at 1 week explains about 56% of the variance in outcome at 6 months. A similar logistic regression was used to predict the likelihood of recovery of hand dexterity at 6 months, assessed using the action research arm test, in patients presenting with flaccid hemiplegia [20]. It was found that if patients failed to reach an arm Fugl-Meyer score of 11 or more by week 4 then they had only a 6% chance of regaining dexterity at 6 months. Notably, this probability did not change over the ensuing 5 months. Thus, there is a process of spontaneous recovery that is maximally expressed in the first 4 weeks post-stroke and then tapers off over 6 months. Several mechanisms are likely for this spontaneous recovery, including restitution of the ischemic penumbra, resol- ution of diaschisis, and brain reorganization. Although some aspects of brain reorganization are probably unique to brain injury, there are large overlaps with development [21,22] and motor learning [23,24]. A recent study in a rat stroke model demonstrates the critical interaction between rehabilitation and spontaneous recovery pro- cesses early after stroke [25]. Rehabilitation initiated 5 days after focal ischemia was much more effective than waiting for 1 month before beginning rehabilitation. This difference correlated with the degree of increased den- dritic complexity and arborization in undamaged motor cortex. A similar time-window effect, albeit longer than in rats, has been shown in patients after stroke, with the greatest gains from rehabilitation occurring in the first 6 months [26]. Improvement with rehabilitation increases with the amount of training and relates mostly to the task practised during therapy, with little generalization to other motor tasks. Thus, recovery related to spontaneous biological processes seems to improve performance across a range of tasks whereas recovery mediated by training, like learn- ing in healthy subjects, is more task-specific. This differ- ence raises the important issue of true recovery versus compensation and how they both relate to motor learning. True recovery means that undamaged brain regions are recruited, which generate commands to the same muscles as were used before the injury. This implies some redundancy in motor cortical areas with unmasking, through training, of pre-existing corticocortical connec- tions [27]. Compensation, in contrast, is the use of alternative muscles to accomplish the task goal. For example, a patient with right arm plegia can compensate by using their left arm. Nevertheless, despite the clear distinction, learning is required for both true recovery and compensation. Experiments in monkeys clearly demon- strate the importance of learning for recovery of function [28,29]. A subtotal lesion confined to a small portion of the representation of one hand resulted in further loss of hand territory in the adjacent, undamaged cortex of adult squirrel monkeys if the hand was not used. Subsequent reaching relied on compensatory proximal movements of 86 Cerebrovascular disease

Copyright �� Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. the elbow and shoulder. Forced retraining of skilled hand use, however, prevented loss of hand territory adjacent to the infarct. In some instances, the hand representations expanded into regions formerly occupied by represen- tations of the elbow and shoulder. This functional reorganization in the undamaged motor cortex was accompanied by behavioral recovery of skilled hand func- tion. These results suggest that, after local damage to the motor cortex, rehabilitative training can shape sub- sequent recovery-related reorganization in the adjacent intact cortex. Critically, cortical changes may only occur with learning of new skills and not just with repetitive use [24]. It is unclear at this time whether simple repetition of a task that was previously well-learned is sufficient to induce significant cortical reorganization or whether patient should be challenged on more difficult tasks. The answer may depend on the amount of salient error information provided (see section below on virtual reality). The ability to compensate for a deficit is also dependent on motor learning, as any right-hander who has tried to write with their left hand quickly realizes. Thus to the degree that all rehabilitation is a form of motor learning, it can occur to promote both true recovery and compen- sation. A recent study of focal cortical ischemia in adult rats suggests that motor improvement is mediated prin- cipally by compensatory mechanisms rather than true recovery. Indeed, some animals developed a compensa- tory movement strategy that was more successful than the one used prior to the lesion [30 ]. Most interestingly, the rate of improvement with training was similar before and after the lesion, suggesting that a similar learning mech- anism was operative with and without injury. Another recent study of focal ischemic brain injury in rats suggests that the undamaged (ipsilateral) hemisphere may be the anatomical substrate for compensatory improvement [31]. These animal studies indicate the benefits of detailed behavioral analysis. Unfortunately, outcome scales commonly used in clinical rehabilitation trials do not have the resolution to distinguish between compen- sation and true recovery. This is a serious limitation for a number of reasons. First, it has been stated that the failure of many recent clinical stroke trials may relate more to the choice of outcome measures rather than to the lack of efficacy of the agent under investigation [32]. Second, inappropriate compensatory strategies may limit recovery after stroke [33]. Third, in order to understand brain changes that occur in response to therapy it is imperative that brain changes due to compensation are not misinterpreted as evidence for reorganization. The application of quantitative move- ment analysis and the motor control framework de- scribed in the previous sections should overcome these limitations and allow accurate assessment of the efficacy of rehabilitation techniques. Rehabilitation methods based on motor learning This section will review five rehabilitation techniques based on motor learning principles. Some target patients with a particular degree of hemiparesis while others are appropriate across the spectrum from mild hemiparesis to hemiplegia. It can be envisaged that patients could be tried on the techniques in combination or graduate from one to another as they improve. Arm ability training: impairment-oriented training for mild hemiparesis This technique was developed for patients with mild hemiparesis [34], who complain of clumsiness and decreased coordination even though they may have normal neurological examinations and arm Fugl-Meyer scores. Deficits may only be apparent with more sensitive kinematic testing [35]. These patients, however, are the most likely to return to work after their stroke and so their deficits, albeit mild, can be devastating, for example, for electricians, hairdressers, or musicians. The arm ability training tasks were chosen based on a factorial analysis of different abilities in healthy subjects: hand grip, finger individuation, arm-hand steadiness, aimed reaching, tracking, and wrist-finger speed. The protocol incor- porates many of the concepts from the motor learning literature in order to maximize retention and generaliz- ation of what is learned during the rehabilitation session. For example, although tasks are practiced repetitively, variability is introduced by varying the difficulty of each of the tasks. A randomized clinical trial showed a benefit of arm ability training compared with standard rehabilita- tion, as assessed by a measure of efficiency of arm func- tion in ADLs [34]. The emphasis of the arm ability training protocol focuses on impairment rather than on disability or quality of life measures is more congruent with neuroscientific findings, which indicate that motor control and motor learning are modular [5]. Constraint-induced movement therapy (CIMT) This technique has garnered a large amount of attention because it has shown that even patients with chronic stroke ( 6 months out) can show meaningful gains (for a recent review, see [36]). The technique has two com- ponents and is usually given over 2 weeks: (i) restraint of the less-affected extremity for 90% of waking hours (ii) massed practice with the affected limb for 6 hours a day using shaping. In patients with chronic hemiparesis, the restraint is conjectured to help patients overcome learned non-use, whereas in patients with acute stroke it can be seen as a way to prevent adoption of compensatory strategies with the unaffected limb. Shaping is a form of operant conditioning whereby performance is consist- ently rewarded ��� essentially the reverse of the mechan- ism by which patients are posited to learn non-use. Learned non-use is based on the idea that the affected Motor learning and stroke recovery Krakauer 87