​IntroductionAcute injury to the central nervous system (CNS) is often followed by some degree of recovery. Scientists and clinicians have been interested in the mechanisms of this recovery for years. Based on observations in animal models of focal CNS injury it is often assumed that several processes jointly referred to as neuroplasticity make a major contribution (see Chapters 13 and 14). Experiments in animal models have demonstrated alterations in cerebral organization that occur after injury are related to recovery [1] . Specifically, focal cortical damage in adult brains renders widespread surviving cortical regions more able to change structure and function in response to afferent signals in a way normally only seen in the developing brain [2]. An increased potential for neuroplasticity will in itself not enhance recovery, but it may increase the impact of learning or training strategies since learning works through mechanisms of experiencedependent plasticity [1]. The reverse, however, is not necessarily true (i.e. plasticity does not imply learning has occurred). The management of patients with incomplete recovery following CNS injury often draws on specific rehabilitation interventions aimed at assisting adaptation to impairment. However, partly because of a growing awareness of the role of neuroplasticity there is an interest in designing therapeutic strategies to promote cerebral reorganization as a way of reducing rather than compensating for impairment. These include incorporating ideas about learning into neurorehabilitation (see Chapter 8) as well as strategies to enhance the potential for plasticity, such as neuropharmacological (see Chapter 18) and noninvasive brain stimulation (see Chapter 17). These developments are clearly very exciting for clinicians. A key part of developing future strategies will involve building an empirical understanding of how the brain responds to injury and how such changes may be manipulated in a way that promotes functional recovery [3] . The investigation of cerebral reorganization after focal brain injury in humans is less well advanced than similar work in animal models, but the goal is to provide biomarkers of recoveryrelated biological processes that are difficult to measure directly in humans [4]. There are clearly greater limitations in studying the human brain, but structural and functional imaging provide opportunities to do so. This chapter will explore how neuroimaging has contributed to understanding the impact of neuroplasticity after CNS injury, and how it might contribute in the future. It will largely concentrate on motor recovery after stroke to illustrate how neuroimaging provides a window onto neuroplasticity after CNS damage, but examples from the study of different types of patients (spinal cord injury) and different domains (language) will be referred to in order to examine how much it is possible to generalize these ideas. Imaging techniques Functional imagingFunctional neuroimaging techniques allow examination of human brain function in vivo. In the context of CNS injury, functional brain imaging provides a way of assessing how focal damage to cortical or subcortical regions alters the way surviving neural networks operate, and how these changes are related to impairment and recovery. Functional imaging of the brain has been carried out with four main techniques: positron emission tomography (PET), functional magnetic resonance imaging (fMRI), electroencephalography (EEG) and magnetoencephalography (MEG). A detailed theoretical background to the techniques is beyond the scope of this chapter. In brief however, both PET and fMRI rely on the assumption that neuronal activity is closely coupled to a local increase in cerebral blood flow (CBF) secondary to an increase in metabolism. PET relies on mapping the distribution of inert, freely diffusible radioactive tracers deposited in tissue as a function of regional perfusion (rCBF). Functional magnetic resonance imaging (MRI) comprises different methods, but the studies described next use blood oxygen leveldependent (BOLD) imaging techniques. During an increase in neuronal activation there is an increase in local CBF, but only a small proportion of the greater amount of oxygen delivered locally to the tissue is used. There is a resultant net increase in the tissue concentration of oxyhaemoglobin and a net reduction in paramagnetic deoxyhaemoglobin in the local capillary bed and draining venules. The magnetic properties of haemoglobin depend on its level of oxygenation so that this change results in an increase in local tissue derived signal intensity on T2*weighted MR images. EEG and MEG on the other hand are techniques that measure the magnetic fields emanating from the scalp, which are created perpendicular to the electrical current (according to Maxwell’s equation) that is created by neuronal activity. EEG systems are cheaper and more readily available than MEG, but MEG has some advantages. EEG signals are strongly degraded by heterogeneity in conductivity within head tissues, but this is far less of a problem in MEG. MEG directly measures neuronal activity and has a temporal resolution in the scale of milliseconds. Studies measuring rCBF with PET are less common now, but MEG studies are on the increase. Structural imaging techniquesIdeally, changes in CNS functional organization should be viewed in the context of the anatomy of the structural damage. However, structural imaging in stroke for example has generally been used to examine the vascular territory involved, without too much consideration of the important functions subserved by the grey and white matter structures that are damaged. Up until recently, there has been no good way to quantify damage to key structures using computerized tomography (CT) or T1and T2weighted MRI. Early attempts at quantification used lesion volume, which is not a good predictor of outcome [5] , or tended to reduce complex 3D anatomical data to single numbers (e.g. percentage of corticospinal tract damaged [6]). However, multivariate approaches to assessing the impact of the pattern of brain damage are now readily available and are being applied in the context of stroke recovery [7, 8]. Imaging motor recovery after strokeCrosssectional studies in chronic strokeThe first functional imaging studies to examine cortical reorganization of the motor system were performed in recovered chronic subcortical stroke patients. These patients were found to have relative overactivation in a number of motorrelated brain regions during the performance of a simple motor task compared to control subjects. In particular, overactivations were seen in brain regions such as dorsolateral premotor cortex (PMd), ventrolateral premotor cortex (PMv), supplementary motor area (SMA), cingulate motor areas (CMA), parietal cortex, and insula cortex [9–12]. A recent metaanalysis on activation data derived from over 50 neuroimaging experiments confirmed that enhanced activity in contralesional primary motor cortex (M1), bilateral ventral premotor cortex and SMA are a highly consistent findings after motor stroke compared to healthy controls for a wide range of hand motor tasks [13]. These findings were initially interpreted as indicating that recruitment of these brain regions, particularly those in the unaffected hemisphere, might be responsible for recovery. However, stroke patients are variable and if one studies patients with a range of late poststroke outcome, results suggest that those with the best outcome have a ‘normal’ activation pattern when compared to normal controls, whereas those with poorer outcome show significant differences. Although care needs to be taken in conducting and interpreting ‘taskrelated’ studies, the differences between stroke patients and healthy controls generally take the form of (i) overactivations in nonprimary motor areas, particularly in the contralesional hemisphere and (ii) shifts in somatotopic representation in primary and possibly nonprimary motor areas [14]. In fact, when the relationship between impairment and regional brain activation was examined for the first time, a negative correlation was found between the magnitude of brain activation in secondary motor areas and outcome [15] (Fig. 15.1). In other words, this result confirmed that those with more impairment were the ones with overactivations described just now. A subsequent study used transcranial magnetic stimulation (TMS) to quantify the ‘functional integrity’ of the corticospinal system to test whether this may be the key variable leading to alterations in patterns of taskrelated activity after stroke. Patients with more corticospinal system damage exhibited less taskrelated activity in ipsilesional M1 (hand area) and greater activity in secondary motor areas in both hemispheres [16]. A similar result was observed in a group of patients with different levels of impairment studied at approximately 10 days post stroke illustrating that lesioninduced reorganization occurs quickly [17]. These results point to a shift away from primary to secondary motor areas with increasing disruption to corticospinal system, presumably because in some patients ipsilesional M1 is less able to influence motor output. However, this is highly likely to depend on the exact pattern of disruption to the descending pathways. The results from similar studies performed in patients with injury occurring to the CNS at a much earlier age (e.g. cerebral palsy) provide similar results. Prominent contralesional activity has been observed, both in premotor and primary motor cortex, with the latter more likely to be recruited in those with larger lesions [18]. As in those with adult stroke, there is variability in motor system organization related to lesion extent and level of impairment. The evolution of cerebral reorganization after strokeCrosssectional studies are simpler to perform, but do not tell us is how this reorganized state evolved from the earliest time after infarction. Two early longitudinal studies with early and late time points demonstrated initial taskrelated overactivations in motorrelated brain regions followed by a reduction over time in patients said to recover fully [19, 20]. A detailed multisession longitudinal fMRI study of patients with infarcts not involving M1 looked at changes in motorrelated brain activity as a function of recovery (rather than time). At approximately 10–14 days after stroke, an initial overactivation was seen in many primary and nonprimary motor regions [21]. As in the chronic setting, this was more extensive when the clinical deficit was greatest (i.e. early after stroke). Improvement in motor performance was associated with a steady decrease in taskrelated activity in these areas (Fig. 15.2) suggesting that successful recovery is associated with a normalization of pathologically enhanced brain activity over time, which has been confirmed by a number of subsequent studies [22–24]. Even earlier changes were examined by a serial fMRI study in which stroke patients with motor impairment were scanned several times in the first 2 weeks poststroke starting within 3 days after symptom onset [23]. In those with only mild impairment, taskrelated activation (movement of the affected hand) was not different from healthy controls. However, in those with more marked impairment, there was a general reduction of cerebral activity in the first 1–3 days after stroke, which increased in both hemispheres over and above that seen in healthy controls over the next 10 days. Four months later, cortical overactivity had returned to levels observed in healthy controls in those with recovery of hand function, similar to earlier longitudinal studies. The early absence of activity is an interesting finding that might represent a real decrease in neural activity or possibly merely reduced BOLD due to neurovascular uncoupling. Intriguingly however, reduced BOLD reactivity has been linked with increased levels of gammaaminobutyric acid (GABA) [25]. This is of particular interest as, the balance between inhibition and excitation in the cortex is thought to be a key mediator of neural plasticity. The temporal pattern of reduced then elevated BOLD might point towards the kinds of alterations in lesioninduced plasticity that evolve over time that are seen in animal models of focal brain injury [1] . Brain reorganization in response to therapeutic interventionsThe studies described so far have examined alterations in organization of cortical motor areas in response to damage (to the corticospinal pathways). There are a number of studies that have looked at the effects of physical therapies (for a review, see [26]). The standard design is to use functional imaging before and after a particular treatment protocol. Most found treatmentassociated increases in ipsilesional hemisphere activity in keeping with the previous longitudinal studies, but others saw a shift in the balance of activation in the opposite direction. The evidence suggests that the contribution of contralesional motor regions varies, but it is not clear what baseline characteristics might predict such shifts. In other words, it is likely individual differences in the anatomy of the damage and time since stroke will determine what topography of therapeutic change is observed. These results are likely to represent the consequences of functionalimprovement rather than the mechanism of action of the treatment itself. Only one study has looked at the differential longitudinal changes in brain reorganization for one form of therapy compared to another [27]. Bilateral arm training with rhythmic auditory cueing (BATRAC) led to significantly higher increases in activation in some ipsilesional motorrelated areas including PMd and SMA than after matched intensity ‘standard’ physiotherapy. From a clinical perspective it was disappointing that there were no overall differences in clinical gains for either group. Although a negative clinical trial, it means that the functional imaging differences are not confounded by different therapeutic gains. In other words, the results here do suggest a possible cerebral mechanism for BACTRAC compared to intense ‘standard’ physiotherapy. In general, functional imaging is unlikely to be useful purely as a marker of clinical improvement, something that is measurable with simple outcome scores. Functional imaging may become a useful marker of the potential for change in damaged brain, and this will be discussed later in the chapter. Is this reorganization functionally relevant? Most longitudinal studies have been performed in those who end up with reasonable recovery and support the importance of regaining normal patterns ofbrain activity. However, the crosssectional studies tell us that not all patients achieve this normalization and those with incomplete recovery can be left with prominent taskrelated activity in secondary motor areas, particularly in contralesional hemisphere. What is the evidence that this pattern of cortical activity during attempted movement is either contributing to or hindering recovery of motor function? Do these distributed cortical motor regions have any direct influence over muscles in recovering limbs? One way to look at this is to measure the coherence between oscillatory signals from both the brain (measured with magnetoencephalography, MEG) and the affected muscles (measured with electromyography, EMG) simultaneously during a simple movement. Corticomuscular coherence here implies some kind of functional coupling between the cortical region and the recovering muscle. In a group of chronic stroke patients, the cortical source of the peak corticomuscular coherence was widely distributed compared to controls [28]. In particular, peak corticomuscular coherence was seen in contralesional hemisphere in a number of patients (Fig. 15.3), implying direct influence over affected muscle activity. Transiently disrupting cortical activity in either ipsilesional or contralesional PMd with TMS usually does not affect healthy volunteers, but can lead to worsening of recovered motor behaviours187 in some chronic subcortical stroke patients [29–31]. The effect is usually dependent on residual impairment. For example, TMS to contralesional PMd is more disruptive in patients with greater impairment [30], whereas TMS to ipsilesional PMd is more disruptive in less impaired patients [29], implying a contralesional shift in balance of functionally relevant activity in those patients withgreater impairment. These findings are in keeping with the functional imaging findings discussed above earlier. Another approach is based on the assumption that activity in brain areas that are functionally involved in producing a specific behaviour, covary with modulation of the task parameters. For example, activity in contralesional sensorimotor and premotor cortices might increase in proportion to the frequency of finger movements in well recovered stroke patients in contrast to control subjects [32]. Another study asked subjects to vary force output, rather than movement rate, and then examined for regional changes in the control of force modulation [33]. In healthy humans increasing force production is associated with linear increases in BOLD signal in contralateral M1 and medial motor regions, implying that they have a functional role in force production [34]. In stroke patients with minimal corticospinal system damage and excellent recovery, the cortical motor system behaved in a way that was similar to younger healthy controls. However, in patients with greater corticospinal system damage, forcerelated signal changes were seen mainly in contralesional dorsolateral premotor cortex, bilateral ventrolateral premotor cortices, and contralesional cerebellum, but not ipsilesional primary motor cortex [33]. A qualitatively similar result was found in healthy volunteers with increasing age suggesting that this ‘reorganization’ might be a generic property of the cortical motor system in response to a variety of insults [35]. In relation to lesioninduced reorganization, not only do premotor cortices become increasingly active during movement as corticospinal system integrity diminishes [16], but also take on a new ‘M1like’ role during modulation of force output, which implies a new and functionally relevant role in motor control. The timing of the taskrelated activity might also be useful in determining function in relation to action. For example, using eventrelated fMRI contralesional M1 activity peaks seconds before ipsilesional M1 in stroke patients, in comparison to controls in whom the opposite relationship is observed [36]. On the other hand, in a different study using the fine temporal resolution of electroencephalography (EEG), contralesional hemisphere activity was detected after the motor response had been made suggesting that it was not related to movement initiation in these patients [37]. Despite its temporal resolution, EEG lacks fine spatial resolution, and so it is not certain which contralesional brain region this result related to; M1 or premotor cortex for example. Others have used directed EEG coherence to investigate whether there is increased the flow of information from the ipsilateral motor cortex following motor stroke [38]. This approach suggested that in stroke patients with residual impairment, the contralesional hemisphere was the main ‘driver’ (at least in the βband activity) for taskrelated flow of information during grip with the affected hand, whereas in recovered patients and controls cortical activity was driven from the ipsilesional (contralateral in controls) sensorimotor cortex. The results described so far indicate that there is some novel contribution to motor control from the contralesional hemisphere after stroke. Some studies have moved their attention to the premotor cortex. At rest, it seems that the influence of contralesional PMd on ipsilesional motor cortex is inhibitory in well recovered patients, but becomes more facilitatory in those with greater clinical impairment [39]. By using concurrent TMSfMRI, it was also possible to examine which brain regions contralesional PMd was influencing. During affected hand movement there was a stronger influence of contralesional PMd on two posterior parts of the ipsilesional sensorimotor cortex [39]. This provides a possible mechanism by which contralesional PMd might exert its statedependent influence over the surviving cortical motor system since ipsilesional sensorimotor cortex is most likely to be able to generate descending motor signals to the spinal cord to support recovered motor function The results presented so far suggest that activity in contralesional hemisphere contributes to motor control after stroke, particularly in more impaired patients. However, an alternative view is that motor areas in the contralesional hemisphere, in particular M1, are pathologically overactive after stroke. There are both TMS [40] and fMRI [41] studies which suggest that in some subcortical stroke patents, contralesional M1 although ‘active’, may exert an abnormally high degree of interhemispheric inhibitory drive towards ipsilesional M1 during attempted voluntary movement of the affected hand. In other words, contralesional M1 overactivity somehow suppresses ipsilesional M1 activity and consequently motor performance and recovery. Others have used this concept to suppress excitability in contralesional M1 using noninvasive brain stimulation, in an attempt to enhance the effect of motor training. There are now many small studies [42]. Although initially positive, the publication bias is gradually being corrected and negative studies are being published [43]. What is likely to emerge is that the anatomical and neurophysiological characteristics of the individual patient will determine if and how it is possible to prime the motor system so that training regimes have more effect. This will allow stratification of approaches based on mechanistic understanding [44]. The anatomical substrates of motor recoveryReorganization of cortical motor systems is most prominent in patients with greatest clinical deficit and presumably with the most significant damage to the descending motor pathways. Clearly, recruitment of secondary motor areas does not get patients back to normal, but the evidence is that in many it is at least supporting what recovered function they have. If so, what are the possible anatomical substrates of this effect? A key determinant of motor recovery is sparing of the fast direct motor pathways from ipsilesional primary motor cortex (M1) to spinal cord motor neurons [45, 46] There is little evidence that ipsilateral projections from motor cortex to forelimbs exist in primates [47] although this does not rule out such a possibility in humans. This makes ipsilateral projections from contralesional M1 a less likely substrate, but what about those from secondary motor areas? In primates, projections from secondary motor areas to spinal cord motor neurons are usually less numerous and less efficient at exciting spinal cord motoneurons than those from M1 [48, 49]. Studies in primates in which layer V (the ‘output’ layer) cortical neurons were stimulated and stimulustriggered averages of electromyographic activity measured from forelimb muscles during a reachtograsp task [50, 51]. The onset latency and magnitude of facilitation effects from premotor areas PMd, PMv, SMA, and dorsal cingulate motor area (CMAd) were significantly longer and weaker than those from M1. Although there was evidence for the first time of a small number of direct projections to spinal cord motoneurons at least as fast as those from M1, from each of the secondary motor areas, the majority are unlikely to have a direct influence. Alternative pathways to spinal cord motoneurons would include via corticocortical connections with ipsilesional M1 or via interneurons in the spinal cord. Finally, it is often cited that secondary motor areas only have meaningful projections to proximal rather than distal muscles. In these studies, proximal muscles were predominantly represented in PMd and PMv but for both SMA and CMAd, facilitation effects were more common in distal compared to proximal muscles. These medial motor areas are almost always ‘overactive’ in stroke patients compared to control subjects Another possibility is that premotor areas are able to send descending motor signals via alternative pathways such as reticulospinal projections to cervical propriospinal premotoneurons [52–54]. These pathways have divergent projections to muscle groups operating at multiple joints [55, 56] which might account for the multijoint ‘associated’ movements such as the synergistic flexion seen when patients with only poor and moderate recovery attempt isolated hand movements [52]. Although some see these synergistic movements as a barrier to further improvements in motor control (towards ‘normal’ patterns of movement), it is likely that these patients don’t in fact have enough of the appropriate anatomical substrate (fast direct contralateral projections from ipsilesional M1) to support ‘normal’ movement. In this context, synergistic movements can contribute to functional improvement. Overall, it is feasible that a number of motor networks acting in parallel could generate an output to the spinal cord necessary for movement, and that damage in one of these networks could be at least partially compensated for by activity in another [57, 58]. A potentially exciting area for development would be if the projection characteristics of these alternative pathways could be altered. For example, there is some evidence from primates that reticulospinal projections to specific muscle groups (e.g. biceps) can be strengthened using the principles of spike timingdependent plasticity [59]. Currently there is no evidence that reticulospinal pathways exert any control over finger extensors, but if plasticitybased interventions could change this, then it would open up a whole new therapeutic area. Imaging language recovery after strokeIn the language domain, functional imaging studies of brain reorganization after stroke have focused largely on patients with anomia, a symptom present in almost all types of aphasia. Many of the functional imaging studies have demonstrated poststroke activity in a right hemisphere homologue of either Broca’s (BA 44/45) or Wernicke’s (BA 22) area [60]. Attempts to find a correlation between the magnitude of right hemisphere activation and recovery of language function were unsuccessful, unlike the equivalent studies in the motor domain [15] suggesting that the story is most likely more complicated than simply switching a function from one hemisphere to the other. This is illustrated by a longitudinal study in which early (within 12 days of stroke) overactivity in right Broca’s area compared to controls correlated with better naming ability. After this early phase the relationship between right Broca’s area activity and naming performance altered with declining activity occurring at a time of continued clinical improvement [61]. In the same study, there was little taskrelated BOLD signal very early after stroke (2 days), but it is not clear whether this was neural in origin or due to neurovascular uncoupling. Interestingly, the same early poststroke reduction in taskrelated BOLD signal has been reported in the motor domain [62]. In keeping with this apparent alteration in the relationship between naming performance and right Broca’s activity, attempted disruption of naming with TMS was more successful in the first 2 weeks after stroke compared to 2 months later [63]. So, as in the motor domain, the role of surviving cortical regions changes with time after stroke Looking beyond Broca’s, early recovery of naming ability is dependent on restoration of perfusion to at least one of three key areas in the dominant hemisphere—BA37 (posterior middle and inferior temporal/fusiform gyrus) and BA 22 (Wernicke’s area) as well as Broca’s area (BA 44/45) [64]. The most important of these for naming is BA 37, with perfusiondiffusion mismatch (i.e. salvageable tissue) in this area predicting good recovery [65]. Recovery of single word auditory comprehension however is most likely seen with reperfusion in BA22. One possibility is that damage to more posterior temporal structures (such as BA 22) can disrupt activity in more anterior superior temporal regions that are usually spared in middle cerebral artery territory strokes [66]. As in the motor domain, there have been studies examining treatment related alterations in activation pattern. The results are rather conflicting, possibly because of variations in patients (lesion anatomy, clinical phenotype) and the task used during scanning [67]. The key point to remember is that changes in activation pattern rarely point to the mechanism of the treatment itself, but rather reflect the behavioural improvement that has taken place, irrespective of which treatment was used. The field of functional imaging and language recovery after stroke is rapidly catching up with its counterpart in the motor domain in terms of numbers of publications. The details of numerous studies have been extensively and recently reviewed elsewhere [67–72]. Imaging distant consequences of spinal cord injuryStudies in animals with spinal cord injury (e.g. transection of the dorsal columns) demonstrate extensive reorganization of sensory inputs into the CNS [73–75]. In humans, functional brain imaging studies of those with spinal cord injury also provide evidence that distant neuronal damage has an impact on organization of the whole sensorimotor system. Studies that have examined brain activity during unaffected hand movements in paraplegic patients have shown a variety of changes [76]. Some, but not all have demonstrated expansion or overrepresentation of one body part in the sensorimotor cortex at the expense of another. The magnitude and topography of cortical reorganization is variable and probably depends on a number of factors, in particular the characteristics of the anatomical damage. To examine these relationships explicitly, a recent study looked at the relationships between structural and functional changes following spinal cord injury [77] and found (i) cortical thickness in sensorimotor areas was reduced in patients with spinal cord injury; (ii) taskrelated brain activation during hand grip was greater in M1 (leg) in spinal cord injury (SCI) subjects with greater cord damage; and (iii) subjects with greater cord damage and greater reduction in tactile sensitivity showed greater brain activation of the face area of left S1 during right median nerve stimulation. Improvements in imaging techniques have recently allowed neurodegeneration in both dorsal and ventral horns as well as white matter above the level of injury to be detected. Ventral horn atrophy was associated with motor outcome, while dorsal horn atrophy was associated with sensory outcomes. White matter integrity was associated with dailylife independence [78]. Overall then, it is likely that variability of brain reorganization is driven by differences in anatomical damage. Failure to account for this in studies with small numbers of subjects is likely to lead to contradictory results. These caveats are of course true in stroke studies too, but more recently these problems have been addressed [13]. Assessing network connectivityMany of the studies described here use a ‘voxelwise’ or regionofinterest approach. In other words, inferences are made about activity in certain parts of the brain independently of others. However, we know that the brain is organized in circuits and that brain regions influence one another. Assessing changes in connectivity within surviving networks is an interesting and biologically plausible way to go. Two terms are often used—functional and effective connectivity. The most important difference between these two analysis approaches is that effective connectivity analyses (e.g. dynamic causal modelling, structural equation modelling,) allows inference to be made about the influence that one brain area exerts over another, i.e. there is directionality in the data [79, 80]. Functional connectivity analyses (e.g. coherence or correlation analyses, graph theory) describes coupling between brain regions, but does not allow one to say that either area is influencing activity in the other (e.g. the coupling may be driven by another separate region) [81]. This is most commonly performed on fMRI data collected at rest, without the performance of a task. Restingstate data is most likely to reflect the consequences of changes in structural connectivity, since no actual task is performed. For example, stronger (functional) connectivity between ipsilesional M1 and other brain areas (i.e. more normal) in the early poststroke phase is associated with better functional recovery 6 months later [82]. In particular, interhemispheric connectivity appears important, with reduced functional connectivity between ipsilesional M1 and contralesional M1 associated with greater motor impairment [83, 84]. Whichever approach is used, it is always important to find a link with behaviour, something that is intrinsically easier to do in pathological states than in healthy controls because of the greater variability in performance. It is also useful to compare techniques (usually in the absence of a ‘gold standard’ metric). For example, Boudrias and colleagues [85] examined the influence of left M1 on right M1 during right hand squeeze, with both TMS and DCMfMRI. The variability in this cohort came from the range of ages rather than pathology. The influence of left M1 on right M1 diminished with advancing age, and importantly, the assessment of interhemispheric inhibition with TMS correlated with that measured with DCMfMRI, thus providing face validity for the DCM approach, at least in the cortical motor system. Dynamic causal modelling of fMRI data has been used to show that effective connectivity between premotor areas and ipsilesional M1 was significantly reduced in the early poststroke stages [41). Another finding was of reduced coupling from ipsilesional SMA and PMd to ipsilesional M1 very early (less than 72 hours) after stroke. In patients who improved the most, these coupling parameters returned towards normal over the first few weeks [62]. Little work has been done on what reorganization of multiple largescale brain networks means, other than reflecting structural (and therefore functional) changes in the poststroke brain. One recent study used taskbased multivariate functional connectivity analysis to derive spatial and temporal information of wholebrain networks during hand movement (as opposed to during rest) [86]. Noticeably, stroke patients did not show a reciprocal default mode network deactivation peak following activation of their motor network, suggesting that allocation of functional resources to motor areas during hand movement may impair their ability to efficiently switch from one network to another. Overall however, the results from such studies have yet to converge in a way that provides convincing insights into network reorganization after CNS damage, but continued careful studies with larger numbers of subjects may lead to further insights. Future applications for neuroimaging in neurorehabilitationSo far, we have considered studies that have examined brain organization at different stages of recovery after CNS injury. Although these findings are likely to reflect changes occurring as a consequence of neuroplasticity, it is not clear that they have led to different ways of thinking about how to treat patients with CNS injury. There are two ways that neuroimaging may contribute more directly to clinical care. Firstly, by helping to predict likely outcomes and secondly, to indicate whether a particular treatment approach might benefit an individual patient. Predicting outcomes with neuroimagingThe most obvious way to use neuroimaging to predict outcome after stroke is to assess CNS structure. Diffusion tensor imaging (DTI) is able to assess integrity of white matter tracts and several studies have demonstrated that greater damage to the corticospinal tract (CST) is associated with more impairment [87], while the arcuate fasciculus is being examined in aphasia [88]. These measures may also be used to predict future outcome. CST integrity measured within three weeks of subcortical stroke correlate with both initial and 6month upper limb impairment [89]. In a separate study, damage to the CST at the posterior limb of the internal capsule (PLIC) 12 hours poststroke correlated well with motor impairment at 30 and 90 days [90]. These measures were superior to lesion volume and baseline clinical scores in their predictive power. TMS is also used to assess CST integrity and when combining it with DTI within 4 weeks of stroke, TMS had higher positive predictive value than DTI for upper limb function 6 months later, while DTI had higher negative predictive value [91]. Stinear and colleagues originally developed the PREP (Predicting REcovery Potential) algorithm for sequentially combining simple clinical, TMS, and DTI measures to predict upper limb outcomes [92]. The PREP2 algorithm however performs just as well if not better than its predecessor but did not require the DTI based information regarding CST integrity. Rather than use univariate summary measures (e.g. lesion volume, CST integrity) for prediction, there is a recognition that multivariate methods are likely to be superior. This was demonstrated explicitly in one study, where multivariate approaches to describing patterns of brain damage were superior to univariate approaches when accounting for variability in upper limb motor function in chronic stroke patients [8] . Furthermore, this study showed that using lesion information from only CST led to less accurate predictions than using information from a wider motor network (Fig. 15.4A The PREP algorithms highlight the need to use multiple sources of information, e.g. clinical and imaging (which could include structural, functional, and electrophysiological measures). Neuroimaging should never be used to predict outcome on its own, since we know that initial impairment is a major predictor of longerterm outcomes. In patients who present with severe impairment however, outcomes are highly variable with some patients doing well and others less well [93–95]. There are surprisingly few studies that have attempted to identify predictive characteristics of these two groups. Some studies suggest that less damage to CST results in better outcome despite an ideally severe presentation [96, 97], but using a multivariate approach to answer the same question (in those with severe presentations, can information from the brain scan be used to predict whether a patient will have either a good or bad outcome) found that using only information from CST did not perform as well as using a more widespread cortical network [98] (Fig. 15.4B). The key concept here is that the difference between good and poor recoverers (while matching for initial impairment) will tell us something about what is important for the recovery process itself. In this case, it suggests that damage to CST may be more important for determining initial motor impairment, but more widespread cortical areas subserving functions important for learning (sustained attention, memory, sensation, etc.) are more likely to be important for recovery. Similar multivariate approaches are being used in the predicting language outcome and recovery (PLORAS) project to predict language outcome after stroke (see Chapter 21) [99, 100]. The potential for such an approach extends to many domains including motor and cognitive outcomes. Using this type of neuroimaging complex biomarker discovery [101] we should be aiming to provide accurate prognostic models allowing accurate goal setting in neurorehabilitation and perhaps more crucially, stratification in clinical trials [44]. Functional MRI data acquired in the first few days after stroke has also been used to try to predict a subsequent change in motor or language performance with some success [102–104]. Other studies have become interested in specific metrics such as interhemispheric connectivity as a predictor of outcome, but again future approaches are more likely to be successful in making accurate predictions by using as much data as possible through multivariate approaches, rather than through data reduction designed to make analysis simpler. Predicting treatment response with neuroimaging.Predicting outcome will be useful for clinical and research stratification, but what a clinician would like to know is what are the chances of a patient responding to a specific intervention. Stinear and colleagues [104] set out to determine whether characterizing the state of the motor system would help in predicting an individual patient’s capacity for further functional improvement at least 6 months poststroke in a subsequent motor practice programme. In an approach similar to the subsequent PREP algorithm [92], TMS, structural MRI, and on this occasion functional MRI were used. In patients with motorevoked potentials (MEPs), meaningful gains with motor practice were still possible 3 years after stroke. The situation in patients without MEPs has always been more difficult to predict in the clinical setting but is often taken as a poor prognostic sign [105]. DTI assessment of CST integrity allowed further stratification into responders and nonresponders. Interestingly, the patients also performed a simple motor task during fMRI, but the results as assessed by the degree of lateralization to one hemisphere or the other did not contribute to the predictive model. Cramer and colleagues set out to determine predictors of clinical improvement during three weeks of robotic rehabilitation therapy [106]. Only percentage CST injury and M1M1 interhemispheric connectivity made significant contributions to the predictive model, which accounted for 44% of variance in treatment induced behavioural gains. In general, it seems that a more normal anatomy allows greater change and that the anatomy of the damage is likely to set a limit on how well individual patents will respond to therapy. These kinds of studies illustrate how multimodal imaging and neurophysiological data could be used to assess the state of the motor system and predict the potential for therapy driven functional improvements. At present, we are not able to tailor therapies to individual patients, but these studies illustrate the way forward. Clearly, there needs to be progression from proofofprinciple to incorporatingpredictive tools into larger trials and there is some evidence that it is possible to attempt this [107]. Understanding spontaneous biological recoverySpontaneous biological recovery refers to the early poststroke period during which lesion induced biological changes in the brain support both repair and enhanced responsiveness to behavioural training. It is often described as a period of enhanced plasticity potential and therefore a window of opportunity for stroke recovery. It is likely that epigenetic mechanisms lead to changes in cortical inhibitory mechanisms and then downstream changes in structural plasticity through altered expression of brain derived neurotrophic factor for example. Not all changes are positive [108], but the balance between mechanisms that might support or hinder recovery is not clear, particularly in humans. To address these questions requires the appropriate biomarker with which to bridge the gap between animal and human accounts of poststroke recovery [109, 110]. Here, a biomarker could be any measure that is an indicator of disease state that can be used clinically as a measure reflecting underlying cellular processes that may be difficult to measure directly in humans, and can be used to predict recovery or treatment response [111]. At a recent international consensus meeting, neuronal oscillations measured in the human brain with MEG were highlighted as a promising biomarker of GABAergic activity and therefore the potential for experience dependent plasticity [112]. MEG is a noninvasive neurophysiological technique that measures the magnetic fields generated by neuronal activity of the brain. Specifically, MEG measures the summation of postsynaptic fields from pyramidal cells [113] with excitatory glutamatergic projections, which are reciprocally connected to interneurons with inhibitory GABAergic projections. MEG signals are therefore dependent on the interaction between inhibition and excitation within cortical microcircuits[114] and can provide the appropriate biomarkers of net inhibitory and excitatory mechanisms in human cortex and differentiate the contribution of phasic and tonic inhibition. Experimental evidence to support this is scarce, but stroke patients with poorer outcomes have increased lowfrequency oscillations [115], similar to those caused by benzodiazepines (a GABAA agonist) and tiagabine (a GABA reuptake inhibitor) [116–118], suggesting prominent inhibitory mechanisms in perilesional cortex may hinder recovery. Reversing these increased lowfrequency oscillations with zolpidem has led to clinical improvements in human stroke cases [119, 120]. Zolpidem acts on α1containing GABAA receptors, but can also increase or reduce α5 subunit dependent tonic inhibitory mechanisms depending on dose [121]. Conversely, increased early poststroke sensorimotor excitability (less βrebound in response to tactile finger stimulation) [122] and sensory map size[123] predict good recovery. Work in humans to date therefore suggests that measurement of neuronal oscillations might serve as an appropriate biomarker of GABAergic activity in humans and that enhanced GABAergic mechanisms after stroke may hinder recovery, but it is not yet ready for use in clinical trials. ConclusionsIn summary, CNS damage leads to reconfiguration of brain networks with some brain regions adopting the characteristics of damaged or disconnected regions. This reorganization varies across patients, but does so in a way that appears to be at least partially predictable. Reorganization of regions and networks is often not successful in returning performance back to preinjury levels—the extent of anatomical damage plays a significant limiting role—but it probably helps an individual to achieve some of their potential level of recovery. The potential for functionally relevant change to occur will depend on a number of other factors beyond the anatomy of the damage, not least the biologic age of the subject and the premorbid state of their based on levels of neurotransmitters and growth factors which are able to influence the ability of the brain to respond to afferent input might be determined by their genetic status [124]. Predicting treatment effects will be based on understanding the interactions between these factors [44]. It is clear that individual differences will have a major influence on how a patient might respond to restorative therapies, and it is in this context that modern neuroimaging (together with neurophysiological) techniques may be able to shed light on brain reorganization after CNS damage in individual subjects. Future work should aim to use these kinds of approaches to determine whether assessment of individual postinjury residual functional architecture can be a major predictor of outcome, opening the way for stratification of patients based on the likely response to an intervention. 

Close window by clicking in grey area >