9 - 25: Diagnostic Approach to Functional Recovery: Functional Magnetic Resonance Imaging after Stroke Open Access

Stroke and other cerebrovascular diseases remain the world's second leading cause of death [1] and stroke is the leading cause for acquired disability in adults, including hemiparesis, dysphasia, neglect or other focal neurological deficits. Recent advances in neuroimaging enable rapid and precise diagnosis and new treatment options have become available for patients with acute ischemic stroke, if diagnosis is made within the first hours after the onset of ischemia [2,3]. However, the majority of patients have either only limited effect or are uneligible for revascularization therapy and long-term rehabilitation remains the most important treatment option. In patients with acute stroke, it is difficult to predict functional recovery and the long-term functional outcome varies from patient to patient [4]. A detailed assessment of lesion location and size with structural magnetic resonance imaging (MRI) is often of limited value in terms of explaining or predicting interindividual differences in long-term recovery because structural MRI provides only little information regarding the potential of the nondamaged brain regions to promote recovery of function [5,6,7].

Here functional MRI (fMRI) comes into the picture because the distributed neural activity of functional brain networks can be readily studied with fMRI at rest and while patients perform a specific task [8]. In healthy individuals, fMRI has proven to be a valuable tool to study functional brain reorganization due to learning and long-term practice [9,10] or associated with brain maturation during childhood and adolescence [11] or healthy aging [12]. In a wide range of diseases, fMRI has been extensively used to study how a given brain disease changes the functional neuro-architecture at the systems level [13,14,15]. In the last 10 years, cross-sectional as well as longitudinal fMRI studies after stroke have provided important insights into changes of the brain in recovery after stroke.

In this chapter, we review the application of fMRI to study the reorganization of functional brain networks after stroke.

When stroke patients undergo fMRI, we measure local changes in regional neural activity using the blood-oxygenation-level-dependent (BOLD) signal [16]. A regional increase in neural activity triggers an increase in local blood perfusion. Under normal physiological conditions, regional oxygen supply increases as a consequence of increasing perfusion, exceeding the local activity-dependent increase in oxygen consumption. Accordingly, an increase in regional neural activity leads to a rise in the local oxyhemoglobin concentration and a decrease in the local concentration of deoxyhemoglobin. The activity-driven reduction of paramagnetic deoxyhemoglobin causes the regional increase in the BOLD signal. Hence, the BOLD signal provides an endogenous contrast which is sensitive to regional changes in neural activity, yet it needs to be borne in mind that the BOLD signal is an indirect (vascular) measure of neural activity which relies on neurovascular coupling [16,17]. This explains why fMRI can identify functional brain networks, which show a temporally correlated BOLD signal increase in response to a stimulus or in relation to an experimental task [18,19].

A given brain function is maintained by the functional integration of neural processing among specialized brain regions. Stroke causes a focal brain lesion, which involves one or more specialized brain regions and their interaction with the remaining nodes of the functional network. In other words, the post-stroke brain is characterized by an altered functional network architecture, one which is less effective as opposed to the intact brain, but which will use its remaining processing capacities to maintain as much as possible functional integrity. The altered neural processing within post-stroke brain networks can be studied with BOLD fMRI which can reveal altered levels of regional brain activation within the network as well as changes in the functional interactions between the remaining network nodes.

In stroke patients, fMRI can either be performed while patients are ‘at rest' (i.e., resting-state fMRI) or while patients are exposed to sensory stimuli (i.e., stimulus-related fMRI) or perform a well-defined task in response to a sensory stimulus (i.e., task-related fMRI). These fMRI techniques have been successfully applied in post-stroke patients to assess functional remodeling of brain networks as reflected by regional changes in neuronal activation and distributed changes in functional brain connectivity. Stimulus-related, task-related and resting-state fMRI capture different aspects of functional reorganization and should be considered as complementary techniques with specific strengths and weaknesses. For resting-state and stimulus-related fMRI, it is not necessary that patients can perform a specific task. This has the advantage that these fMRI examinations are feasible even in severely affected stroke patients and can be used to study spontaneous fluctuations in regional BOLD levels (i.e., resting-state fMRI) or changes in regional BOLD signal driven by ‘passive' sensory stimulation (i.e., stimulus-related fMRI).

Resting-state fMRI can be used to study alterations in functional brain connectivity after stroke because the low-frequency (<0.1 Hz) BOLD signal fluctuations at rest are temporally correlated in functional brain networks [8,20]. The resting-state BOLD signal correlations are sensitive to head movements [21]. Moreover, comprehensive filtering should be applied because physiological noise from cardiac and respiratory cycles causes BOLD signal changes resembling those observed in resting-state fMRI [22]. A resting-state fMRI time series can reveal functional connectivity of several functional brain networks, including the so-called default mode network and the motor network [8]. Studies on healthy resting subjects have shown that brain networks which display correlated resting-state activity strongly overlap with the topography of brain networks as identified by task-related fMRI [8].

In contrast, task-related fMRI offers the possibility to identify changes in the task-specific activation pattern after stroke and to examine how task-specific activation patterns dynamically change during the course of recovery. Task-related fMRI studies offer better possibilities to directly relate specific activity or connectivity changes in the relevant brain networks to the degree of functional impairment and to recovery of a specific brain function such as hand paresis, aphasia or neglect. In summary, resting-state, stimulus-related and task-related fMRI measure different aspects of functional integration and therefore, should be used as complementary approaches when assessing functional brain reorganization after stroke.

The above-mentioned fMRI approaches can be combined with an intervention. For instance, focal transcranial brain stimulation might be combined with fMRI to experimentally manipulate the function of one or more of the nonaffected cortical areas [23,24]. This combined brain stimulation-fMRI approach is particularly interesting if one wishes to test the functional relevance of a specific cortical area for recovery of a specific brain function. Another interventional approach is to map distributed changes in the BOLD signal in response to an acute pharmacological intervention compared to placebo. Pharmacological fMRI might be useful to examine how the pharmacological manipulation of a specific neurotransmitter or ion channel alters the functional integration within brain networks and hereby promotes recovery of function [25].

Stroke patients frequently undergo MRI as part of their diagnostic workup. fMRI carries the same contraindications as conventional MRI scans, that is metal implants, claustrophobia etc. Artifacts induced by head movements remain a limiting problem in the acute phase [26]. Here prospective motion correction of head movements using data from optical tracking systems might significantly help to reduce motion artifacts in future studies [27].

As pointed out above, resting-state fMRI is suited for patients with any neurological deficit of any severity as there is no task to perform. The only practical limitation might be related to spontaneous body movements during the resting-state fMRI session. The estimation of resting-state functional connectivity becomes more reliable the more time points (i.e., brain volumes) are acquired during a single resting state, because resting-state connectivity describes the temporal correlation of spontaneous BOLD signal fluctuations within functional brain networks. Van Dijk et al. [28] reported that a scanning session of 5 min is sufficient to acquire reliable resting-state fMRI data with a TR of 2.5 s and a spatial resolution of 2-3 mm. Usually, a resting-state fMRI session lasts between 5 and 10 min which allows resting-state fMRI to be incorporated into existing clinical MRI protocols for stroke.

Stimulus-related fMRI is also relatively easy to establish in a clinical setting and might be used even in patients with a severe deficit. For instance, an auditory language comprehension paradigm with alternating periods during which speech or reversed speech is presented via headphones can be applied in patients with acute aphasia [29]. For task-related fMRI studies, selection of the experimental task is constrained to tasks the patient is able to perform [30,31]. Usually, the experimental task should be as simple as possible, but still specifically activate the neural networks of interest (e.g. the language system in dysphasia or the motor system in motor stroke). If a task is used that is too difficult for the patients, task-related fMRI will inevitably reveal an alteration of task-related brain activity in stroke patients relative to healthy controls, but it will be impossible to interpret the functional significance of the change in brain activity. It is advisable to match task performance in terms of effort. For instance, in patients with hand paresis caused by motor stroke, one might use a grip force task in which patients have to produce a force level relative to their individual maximum grip force rather than a fixed grip force level that is identical for all subjects [32]. Task-based fMRI paradigms may consist of interchanging periods of task performance and rest (i.e., blocked design) or intermingled trials (i.e., event-related design). Blocked fMRI designs are usually preferred as they reveal more robust task-related activations. Regardless of which task patients perform during fMRI, task performance should be monitored as closely as possible. Measures of task performance should be obtained and used as external variables to inform the fMRI data analysis.

Depending on the complexity of the fMRI design, stimulus- and task-related fMRI sessions usually last between 5 and 15 min. Once the equipment for stimulus presentation and performance monitoring is established in the fMRI environment, stimulus- and task-related fMRI studies can be performed in a clinical setting, even shortly after stroke onset [29,32]. However, task-related fMRI is logistically demanding, especially in the acute stage after stroke, because patients need to be familiarized with the task before scanning and task-related fMRI requires a nonroutine setup (i.e., stimulus presentation and synchronization with fMRI data acquisition as well as task performance monitoring). Figure 1 illustrates changes in activation pattern in a longitudinal task-based fMRI study of patients with aphasia after stroke compared with healthy controls.

Early fMRI studies in stroke have focused on changes in regional brain activation rather than assessing the functional interaction between the activated brain regions. In recent years, fMRI has been successfully applied in analyses of functional and effective connectivity on fMRI data acquired in post-stroke patients to investigate how the focal brain lesion caused by stroke alters the interaction between the nonaffected areas of a functional network and how changes in connectivity relate to functional impairment and recovery [33].

Interactions between functionally specialized areas can be described in terms of functional or effective connectivity [34,35]. Studies of patterns of ‘functional connectivity' are based on coherence or correlation of signal changes among cortical regions and thus, merely reflect statistical dependencies among brain regions. It should be noted that functional connectivity neither makes any explicit reference to specific directional effects or causal interactions between brain areas nor refers to an underlying structural network model. Functional brain connectivity can be estimated in a variety of ways, for example through computing cross-correlations in the time domain or mutual information [35].

‘Effective connectivity' describes causal interactions among distinct neural nodes. In contrast to functional connectivity, effective connectivity specifies directional effects of one neural element of a brain network over another. Functional connectivity patterns are often extracted from fMRI time series that have been acquired in a (task-free) resting state, whereas effective connectivity is usually inferred from task-based fMRI time series [36]. It should be mentioned that BOLD fMRI is not the only method with which one can study functional and effective connectivity. Other neuroimaging modalities such as electroencephalography and magneto-encephalography can also be used to analyze functional or effective connectivity patterns in the human brain. The techniques used for extracting effective connectivity patterns from a BOLD fMRI time series are either based on a prespecified anatomical model (e.g. dynamic causal modeling) or largely model free (e.g. Granger causality [37]). Especially dynamic causal modeling has been successfully applied to fMRI data in subcortical motor stroke to reveal impaired integration within the motor network [38].

As pointed out above, fMRI studies in stroke patients can broadly be divided into task-related and resting-state fMRI studies and the focus of interest can be on the distribution of activation within a network (‘activation pattern') or on changes in functional integration (‘connectivity pattern'). Due to space restrictions, we only review key fMRI studies of stroke-induced reorganization in the motor system. However, we wish to stress that fMRI has been successfully used to study functional reorganization in post-stroke patients presenting with other neurological deficits such as spatial neglect or dysphasia [29,39,40,41].

fMRI studies on motor stroke have focused on recovery of motor hand function by mapping task-related brain activation during whole hand grasp or finger movements. Since most studies were restricted to patients with subcortical lesions, little is known about motor reorganization following cortical or corticosubcortical stroke. This complicates the comparison with animal studies, which have almost exclusively examined motor reorganization triggered by cortical stroke lesions [42,43].

After subcortical motor stroke, patients commonly show overactivations in secondary motor areas, including the dorsal premotor cortex (PMd), ventral premotor cortex, supplementary motor area (SMA) and cingulate motor area in the affected and unaffected hemisphere as well as the contralesional primary motor cortex. As a rule of thumb, activation of secondary cortical motor areas is more pronounced in patients with poorer outcome, suggesting stronger recruitment in those patients ‘with greatest need' [44]. Several studies associate persistent overactivation negatively with function and recovery, while refocusing and normalization of activation patterns point towards new network organization akin to the network before stroke and correlate with better outcome [23,45,46,47]. Yet other studies showed a persistence of movement related overactivity, even after nearly full functional recovery [48,49,50]. This divergence may be explained through differences in the degree of impairment, time after stroke and the imaging task.

Using the coordinate-based activation likelihood estimation (ALE) method, a recent meta-analysis of 36 task-related neuroimaging studies specifically addressed the question which motor activation patterns are consistent across studies [51]. Increased activation in contralesional M1 and bilateral premotor areas was a highly consistent finding after motor stroke despite considerable differences among studies in terms of fMRI tasks and motor impairment levels. With respect to motor outcome, the recruitment of the original functional network rather than on contralesional activity was associated with good motor recovery. However, Ward et al. [52] showed that the ipsilateral PMd showed a linear increase in task-related activation as a function of handgrip force in chronic stroke patients with significant impairment, but not in chronic stroke patients with good outcome or in healthy controls. This observation suggests that activity in ipsilateral motor cortical areas such as PMd might also be functionally relevant in patients with poor outcome.

Another important question which can be addressed with fMRI is how a focal stroke lesion affects the functional integration among brain regions forming a functional network. In patients with subcortical motor stroke, a task-related fMRI study employed dynamic causal modeling to demonstrate a reduction in intrinsic connectivity between ipsilesional SMA and ipsilesional primary motor hand area very early after stroke (less than 72 h after symptom onset) [53]. The reduction in positive coupling between ipsilesional SMA and primary motor hand area was found during finger movements with the affected and unaffected hand and correlated with individual motor impairment [53]. While these data point to impaired information flow between ipsilesional brain regions, other fMRI studies suggest that changes in interhemispheric connectivity between homologous motor regions might be associated with poor recovery. In patients with subcortical motor stroke, the primary motor cortices express an abnormal pattern of interhemispheric effective connectivity with the contralesional motor cortex exerting an abnormal inhibitory drive towards the ipsilesional M1 during movements with the affected hand [53]. Likewise, a recent resting-state fMRI study showed that a loss in homologous interhemispheric functional connectivity in the somatomotor network predicted individual impairment of upper extremity function in 23 patients with acute stroke [54]. Finally, it should be noted that interleaving transcranial magnetic stimulation (TMS) with fMRI could be used to test changes in effective connectivity in specific cortical connections by targeting a specific cortical area with focal TMS [55]. This possibility was exploited in a recent concurrent TMS-fMRI study [23], in which short bursts of TMS were applied to the contralesional PMd during fMRI. Interleaved TMS-fMRI revealed that the contralesional PMd had a stronger influence on the ipsilesional sensorimotor cortex when patients moved the affected hand in patients with greater clinical impairment [23].

It needs to be borne in mind that the results obtained with fMRI are correlative in nature. While fMRI can be used to demonstrate a task-related overactivation of a cortical area or a change in corticocortical connectivity, this does not imply that these changes are functionally relevant. In contrast to fMRI, focal TMS is an interventional method that can transiently interfere with ongoing neuronal activity in the stimulated brain area. The ‘interventional nature' of TMS opens up unique possibilities to probe the functional relevance of a change in regional activity as revealed by fMRI [55]. For instance, the observation that the PMd is overactive during motor tasks in patients with motor stroke does not prove that this constitutes a functionally mechanism for recovery. However, this was demonstrated by disrupting neural processing in the PMd with focal TMS: focal TMS given to ipsilesional [56] or contralesional PMd [57] affected the performance of a simple motor task in patients with chronic stroke but not in healthy controls. The disruptive effect of TMS to contralesional PMd was found to be stronger in patients who showed greater motor impairment [57].

Cross-sectional fMRI studies provide a snapshot of the change in functional neuroarchitecture at a given time point after stroke, but they are not suited to clarify how functional reorganization dynamically evolves after stroke. Using a grip force task, a longitudinal fMRI study of motor recovery after subcortical stroke found an initial overactivation in many primary and secondary motor regions when patients performed manual motor tasks with their affected hand [32]. This was followed by a gradual focusing of movement-related activation, in a way that is typical for motor skill learning in healthy individuals. The degree to which the activity pattern shrunk towards a normal (minimized) activity pattern correlated with long-term motor recovery. Poor recovery was associated with a persistence of task-related overactivation, whereas the activation pattern normalized in patients with good recovery. Another longitudinal fMRI study also found dynamic changes in task-related activation during the first 2 weeks after subcortical motor stroke, which depended upon the degree of initial motor impairment. In that study, bilateral increases of activity in the primary motor cortex, lateral premotor cortex, and SMA correlated with short-term motor recovery [47].

With respect to impaired connectivity, a longitudinal fMRI study used dynamic causal modeling to show reduced positive coupling of ipsilesional SMA and lateral premotor cortex with the ipsilesional primary motor hand area in the acute stage (≤72 h [58]). This ipsilesional premotor-to-motor coupling increased over time and the increase was associated with better recovery. The same study also found dynamic changes in interhemispheric effective connectivity. In the acute stage, the negative coupling strength from ipsilesional motor areas to the contralesional primary motor hand region was attenuated. The subacute stage was characterized by a positive influence of the contralesional primary motor cortex on ipsilesional primary motor cortex. The negative coupling between ipsilesional areas and the contralesional primary motor hand area M1 normalized over time. Interestingly, poor recovery in the chronic stage was associated with enhanced negative coupling from the contralesional to ipsilesional primary motor cortex.

Repeated resting-state fMRI measurements also revealed dynamic changes in functional connectivity in the motor network after stroke [59]. Patients with subcortical motor stroke underwent 5 resting-state fMRI measurements in the first year after stroke. A functional connectivity matrix among 21 motor brain regions was constructed and analyzed using graph-theoretical approaches. Overall, the topology of the motor execution network gradually became more random over time, indicating a less efficient network topology. The ipsilesional primary motor area and contralesional cerebellum showed increased regional centralities within the network, whereas the ipsilesional cerebellum showed decreased regional centrality over time. These topological connectivity measures correlated with different clinical outcome measures.

Together, the fMRI studies on patients with motor stroke consistently show that a focal stroke lesion typically affects neural integration of the entire motor system, and clearly emphasize the relevance of a network-based neuroimaging approach to understand functional brain reorganization after stroke [60]. Yet the reported findings are partially conflicting and the functional relevance of specific connectivity changes for motor recovery, for instance the relevance of interhemispheric versus (ipsilesional) intrahemispheric connectivity changes, remains to be clarified.

In recent years, several groups have started to address the question whether fMRI can contribute to predict recovery of a focal neurological impairment in a single patient. In patients with acute ischemic stroke, there are well-established clinical variables such as age and NIHSS score within 6 h of symptom onset that help to predict 3 months' survival and independence (Barthel index ≥95) [61]. However, these clinical variables are too nonspecific to enable prediction of recovery with respect to specific deficits such as upper limb or language improvement. Accurate prediction of upper limb or language recovery might inform rehabilitation planning and assist clinicians and patients in realistic goal setting. This has prompted some researchers to measure brain activation with fMRI in the first few days after stroke in order to test whether the fMRI data contain some information that predicts subsequent improvement in manual motor function [62,63] or language abilities [64].

Marshall et al. [62] studied 23 patients with fMRI within the first days after acute stroke. During fMRI, patients performed a simple repetitive hand closure task in synchrony with a 1-Hz metronome click alternating with rest. A multivariate analysis yielded a correlation between brain activation and change in Fugl-Meyer score over the next 3 months as indicator of motor recovery. Additionally, voxel-based univariate statistical analysis revealed 2 small clusters in the ipsilesional postcentral gyrus and cingulate cortex where initial task-related activation correlated with subsequent recovery. Using the same motor task, the same group subsequently reported that the distributed fMRI activation pattern in combination with the initial Fugl-Meyer score improved the prediction of upper limb recovery in patients with severe initial upper limb paresis (predictive explanation: 47%) as opposed to the Fugl-Meyer score alone (predictive explanation: 16%) [63]. However, this improvement in outcome prediction did not reach significance. In patients with mild initial paresis, the clinical predictive variable (i.e., initial Fugl-Meyer score) already predicted motor recovery with a very high accuracy (predictive explanation: 96%) [63]. It should be mentioned that other mapping techniques such as TMS and diffusion tensor imaging might also help prediction of upper limb recovery [65]. Therefore, the additional value of task-related fMRI needs to be determined.

Saur et al. [64] applied a multivariate machine learning approach (i.e., support vector machine) in 21 stroke patients with moderate or severe aphasia to show that language fMRI data obtained early after stroke contain substantial predictive information about subsequent recovery of language function. In that study, fMRI was acquired during an auditory comprehension paradigm 2 weeks after stroke. Outcome after 6 months was either classified as good or bad. In addition to the fMRI activation pattern, age and initial language deficit were included in the predictive model. Of note, the classification algorithm allowed for the possibility that within a given voxel, the same outcome could be coded by either an increase or decrease in activity during the language comprehension task. A bad outcome could be coded by both high and low activation, while medium activation would then predict a good outcome or vice versa. The multivariate machine learning approach correctly separated patients with good and bad language performance 6 months after stroke in 3/4 of patients when classification was only based on the fMRI activation pattern. Classification accuracy further improved to 86% when age and the initial language impairment were included for classification. A comparable accuracy was reached for the relative language improvement when fMRI data were restricted to a region of interest in the right frontal gyrus.

Together, these initial studies support the idea that the task-related fMRI activation pattern performed early after stroke might be used to predict the recovery of specific brain functions. The same may hold true for fMRI-based connectivity patterns, both at rest or during a specific task. For instance, interhemispheric functional connectivity as revealed by resting-state MRI in combination with clinical variables may constitute a useful predictive marker for recovery [60].

Many prospective fMRI studies have shown that therapeutic interventions can induce significant changes in task-related activation and connectivity [33,66]. Even short-term interventions such as a single session of TMS [24,67] or a single pharmacological challenge [25] can cause consistent shifts in brain activation and connectivity. Other fMRI studies have reported changes in task-related activation or connectivity after long-term training of motor or language functions that were correlated with training-induced improvement [57,66,68]. For instance, James et al. [68] used resting-state fMRI to investigate the impact of 3 weeks of intensive upper limb rehabilitation therapy on interhemispheric connectivity between the ipsi- and contralesional PMd. Structural equation modeling of the fMRI data yielded a stronger influence of ipsilesional PMd on its contralesional homologue after therapy.

In summary, the majority of studies showed that remodeling of cortical functions is possible even years after stroke, including homologous ipsilesional and contralesional regions. Usually, therapy-induced reorganization occurred within the pre-existing functional brain network rather than recruiting new brain regions that belong to other functional brain networks.

It is recommended to screen for macrovascular abnormalities in the arteries supplying the brain. The presence of uni- or bilateral stenosis or occlusion in the major intra- or extracranial arteries supplying the brain may hamper downstream perfusion in specific vascular territories and alter the temporal dynamics of the BOLD response in specific vascular territories. This might be of relevance when assessing stimulus-induced or task-related changes in regional brain activity or connectivity.

Another phenomenon which needs to be considered when performing fMRI in the acute phase of stroke is diaschisis. Diaschisis is defined as a dysfunction of preserved cortical brain regions that are remote but functionally connected to an acutely damaged brain area. This dysfunction leads to a depression of regional neuronal metabolism and cerebral blood flow and thus, will affect the regional activation and interregional connectivity patterns as revealed by fMRI in acute stroke patients. It is therefore difficult to determine how much the patient's initial deficit and subsequent recovery can be attributed to the focal brain damage or secondary diaschisis-related phenomena [41].

Another intrinsic problem relates to the heterogeneity of patient populations. Patients usually present with a combination of neurological deficits, which vary in magnitude from patient to patient. Deficits such as aphasia, hemianopsia or neglect might affect task performance in other tasks because patients might not understand the instruction or fail to appropriately perceive the stimulus that instructs the task. The same applies to the interindividual variability of the localization and extent of brain lesions. Here voxelwise lesion-behavior mapping might help to figure out which areas of the brain need to be damaged by stroke to result in a specific neurological deficit [69]. To this end, the infarcted brain area is delineated manually and a binary brain map is generated containing either affected or preserved voxels for each patient. Patients are further classified according to the presence or absence of a specific neurological deficit. By pooling the data of a large group of stroke patients, a statistical brain map can be generated which identifies those clusters of voxels where local brain damage is most consistently associated with the symptom of interest. Another strategy to cope with the large interpatient variation in terms of anatomical lesion and clinical deficits is to apply more stringent inclusion criteria by including only patients with a prespecified neurological deficit or stroke location. While this increases the comparability among patients, the results of such fMRI studies cannot easily be generalized to the general stroke population. As pointed out above, the majority of fMRI studies on motor stroke have only included patients with subcortical stroke. Therefore, the fMRI results obtained in this subgroup of patients might tell little about motor reorganization that occurs in stroke patients with cortical involvement.

Finally, the potential for functional reorganization critically depends on the nondamaged brain regions and connections that offer the anatomical substrate supporting functional recovery. Therefore, thorough structural mapping of the nondamaged brain will greatly facilitate the interpretation of the fMRI data. In this context, diffusion-sensitive MRI techniques are of great value as they not only allow to define the infarcted area, but also to test whether and how much the major fiber tracts in the cerebral white matter are still intact after stroke. Here diffusion MRI-based tractography can be used to find out which corticocortical or corticosubcortical routes are still available for compensation after stroke-induced focal brain damage [70].

Within the last decade, the use of fMRI in patients with stroke has substantially advanced our understanding of the mechanisms underlying functional brain reorganization in response to a focal brain lesion. There is also some evidence to suggest that fMRI in the acute phase might have some potential to predict recovery. It also appears possible that the results obtained with fMRI will inspire the development of new rehabilitation strategies and assist the planning of future intervention trials. However, the clinical use of fMRI in post-stroke patients is still in its infancy and the establishment of clinically feasible fMRI applications remains a challenge for translational research.

Previous fMRI work in stroke has been confined to small-scale single-center studies and most studies were designed as proof-of-principle studies. Future studies should aim at investigating larger patient cohorts and should include a broader range of stroke patients in terms of lesion location. This will facilitate the generalization of the results and help to identify subgroups of patients that show distinct patterns of functional reorganization. Further, the isolated use of fMRI to study functional reorganization after strokes has clear limitations. We anticipate that a multimodal assessment of functional reorganization that combines different methods, such as fMRI, diffusion MRI and TMS, but also magnetic resonance spectroscopy or electroencephalography will offer a deeper understanding of the mechanisms underlying post-stroke reorganization and its functional relevance. Longitudinal studies starting already few days after stroke are preferable to cross-sectional studies in the chronic stage because they can unravel the spatiotemporal dynamics of recovery. Finally, fMRI research of post-stroke recovery is mainly limited to academic neuroscience centers. It remains a challenge to implement the fMRI approach into nonacademic community hospitals where the majority of patients are treated [71]. This requires the establishment of simple fMRI protocols and automated analysis pipelines that can be implemented as clinical routine. All these considerations need to be taken into account if one wants to foster the clinical use of fMRI in post-stroke patients. This requires a close interaction between ‘Imaging Neuroscience' and ‘Clinical Neurology' to fully realize the potential of fMRI as a means of monitoring the efficacy of therapeutic interventions and stratifying patients based on the likely response to a therapeutic intervention.

Hartwig R. Siebner was supported by a Grant of Excellence sponsored by the Lundbeck Foundation, Mapping, Modulation & Modelling the Control of Actions (ContAct) [R59 A5399].