The Neurofunctional Architecture of Motor Imagery

The remarkable ability of the human brain to mentally simulate sensations and actions has become a major focus of research during the last three decades. Such mental simulation process is called mental imagery and resembles perception, but without the immediate actual corre‐ sponding sensory input. Typically, mental imagery is often referred to as the experience of seeing with the mind's eye, hearing with the mind's ear. In other words, mental images could be generated from any sensorial system. Imagery experience is usually understood as evoca‐ tion, copy, or reconstruction of actual perceptual experience from the long term memory systems, which is first recalled and then internally reproduced within working memory. At other times, participants may also anticipate possible or forthcoming events through imagery as a consequence of what they are expecting. A key feature of this simulation process is that it gives the ability to mentally rehearse motor acts without overt body movement, which is known as the motor imagery (MI) experience. Practically, MI is more often used when the whole human body or body segments are involved, hence requiring participants to imagine the body as the generator of acting forces and not only the effects of these forces on the external world. In this latter case, the information from the body is mainly used to build the mental image of the movement, e.g. tactile or any kind of proprioceptive data, from the muscles, the joints or the inner ear. In the wealth of motor learning literature, it is now well-established that MI is a valuable complement to physical practice to enhance cognitive and motor performance (Driskell et al., 1994; Guillot & Collet, 2008), as well as to promote motor recovery (Sharma, Pomeroy, & Baron, 2006; de Vries & Mulder, 2007; Di Rienzo et al., 2014a). Interestingly, different types of MI can be described and easily combined. While people commonly report using visual imagery (either through a first-person or third-person perspective), auditory imagery, and kinaesthet‐ ic imagery (Kosslyn et al., 1990), each perceptual modality can be accompanied by imagery. On the other hand, and with reference to perception, mental imagery combines several cues from


Introduction
The remarkable ability of the human brain to mentally simulate sensations and actions has become a major focus of research during the last three decades. Such mental simulation process is called mental imagery and resembles perception, but without the immediate actual corresponding sensory input. Typically, mental imagery is often referred to as the experience of seeing with the mind's eye, hearing with the mind's ear. In other words, mental images could be generated from any sensorial system. Imagery experience is usually understood as evocation, copy, or reconstruction of actual perceptual experience from the long term memory systems, which is first recalled and then internally reproduced within working memory. At other times, participants may also anticipate possible or forthcoming events through imagery as a consequence of what they are expecting. A key feature of this simulation process is that it gives the ability to mentally rehearse motor acts without overt body movement, which is known as the motor imagery (MI) experience. Practically, MI is more often used when the whole human body or body segments are involved, hence requiring participants to imagine the body as the generator of acting forces and not only the effects of these forces on the external world. In this latter case, the information from the body is mainly used to build the mental image of the movement, e.g. tactile or any kind of proprioceptive data, from the muscles, the joints or the inner ear. In the wealth of motor learning literature, it is now well-established that MI is a valuable complement to physical practice to enhance cognitive and motor performance (Driskell et al., 1994;, as well as to promote motor recovery (Sharma, Pomeroy, & Baron, 2006;de Vries & Mulder, 2007;Di Rienzo et al., 2014a). Interestingly, different types of MI can be described and easily combined. While people commonly report using visual imagery (either through a first-person or third-person perspective), auditory imagery, and kinaesthetic imagery (Kosslyn et al., 1990), each perceptual modality can be accompanied by imagery. On the other hand, and with reference to perception, mental imagery combines several cues from

Using fMRI to assess brain activity during motor imagery
The most common aim of fMRI data analysis is to investigate the neural processes mediating higher cognitive functions, and detect correlations between brain activations and the task the participants have to perform during the scan. Since the early 1980s, fMRI has come to dominate the MI-related brain mapping research. During typical fMRI image acquisitions, participants are asked to alternatively perform a movement, either physically or mentally. These experimental conditions are also compared with a rest/control condition.
There are two main types of experimental design in fMRI: the block design and the eventrelated design. The block design paradigm consists of several epochs representing experimental conditions, during which stimuli are presented and actual execution or MI are required from participants, which are interleaved with time blocks of rest. Event-related designs rather associate brain processes with discrete events, with stimulus events being randomly presented one at a time, and being separated by an inter-stimulus interval of a pre-determined length. Habituation effects can be reduced, and stimuli are usually presented either tens of seconds apart or at a faster rate (e.g., every second). It should be noted that event-related designs are more sensitive to the details of the hemodynamic response model use. Figure 1 provides a schematic illustration of these two experimental designs when comparing the neural networks mediating MI and actual execution of the same movement.
There is not a universal procedure yet to acquire and process functional data and many researchers independently developed their methods and analyses, using different software.
Practically, the most commonly technique used to map brain activity is based on the Blood Oxygenation Level Dependent (BOLD) effect, measuring regional differences in hemoglobin oxygenation. Therefore, a considerable amount of experimental studies investigated BOLD changes during MI. They primarily aimed at comparing the neuronal activations recorded during MI and actual practice of the same movement. For example, Grèzes and Decety (2001) provided evidence of a functional equivalence between intending, simulating, observing, and performing an action, as shown by the great deal of overlap among the corresponding neural substrates. Interestingly, brain activations recorded during MI do however not exactly match those of actual movement execution (Lotze & Halsband, 2006). Some differences also appear when comparing different types of MI or with regard to the individual features of the sample of participants. Put simply, fMRI allows the measurement of different neuronal activity in participants with high vs. low MI ability , as well as differences in the patterns of cerebral activity related to the degree of expertise of the participants with the motor task (Lotze et al., 2003;Ross et al., 2003;Milton et al., 2007Milton et al., , 2008. Based on such evidence, fMRI offers a way to evidence the MI experience . Finally, recent fMRI data suggested that different types of imagery elicit different patterns of neural activation (Solodkin et al., 2004;Jackson et al., 2006;Guillot et al., 2009;Lorey et al., 2009), while others nicely demonstrated, through positron emission tomography (PET), that the cerebral plasticity occurring after physical practice was reflected during MI (Lafleur et al., 2002;Jackson et al., 2003), hence reinforcing the principle of functional equivalence between MI and physical performance of the same task. Before we consider these issues and review the main related findings in the next section, an important methodological question remains to be addressed. Specifically, what are the additional benefits provided by fMRI to the study of MI compared to other neuroimaging and electrophysiological techniques, and, in turn, what are the limitations of fMRI recordings?
A first advantage of fMRI, compared to PET, is its non-invasive nature. There is no radiation exposure, and no need for injection of radioactive materials with fMRI. Another strength is its high spatial resolution allowing the measurement of brain activity from deep structures. Unlike electrophysiological techniques such as electroencephalography (EEG) or even Brief 'rest' periods can also be included. In the event-related design, each experimental condition is presented at random (orange and red vertical bars and vertical arrows).
The Neurofunctional Architecture of Motor Imagery http://dx.doi.org/10.5772/30961 magnetoencephalography (MEG) which are somewhat biased towards the cortical surface, fMRI records signal from all regions of the brain, including subcortical structures. This is of particular importance when pinpointing the neural correlates of MI as activation of cerebellum and basal ganglia is commonly reported. As every technique, however, fMRI has also some methodological disadvantages or weaknesses for the study of MI (e.g., Dietrich, 2008). A first limit is related to the causality between brain activation and the task the participants perform in the scanner. Although the baseline activity is subtracted from that recorded during the task and the result is averaged for all participants, the causality between activated structures and characteristics of the task is not straightforward. In other words, some brain structures could remain activated although this activity is not directly related to the task. Such argument for the causality however also holds for other brain imaging techniques. Second, the temporal resolution, of about several seconds must also be considered, as fMRI cannot track the rapid temporal dynamics of the functioning brain compared to high temporal resolution methods, e.g. EEG or MEG. Neuromodulatory effects of some cognitive functions such as attention and memory are also likely to affect the spatiotemporal resolution of the signal as they can affect large masses of cells, and potentially induce larger changes in the fMRI signal than the sensory signals themselves (Logothetis, 2008). Fourth, regarding data interpretation, advanced statistical methods are required to identify active voxels, while modifications of the brain activity might be due either to an excitatory or inhibitory influence. Finally, the position in the scanner and the limitation of body movements make it difficult to investigate complex motor skills. Indeed, the participant is lying in supine position in the scanner, and then performs both actual execution and mental representation of the same movement. However, in such situation, only simple movement of fingers, hands, feet, or tongue, or more complex movements of limited amplitude can actually be studied.
As reported by Logothetis (2008), and despite the shortcomings mentioned above, fMRI remains probably the best current available technique for providing insights into brain function and formulating interesting testable hypotheses. Logothetis also stated that some limitations of fMRI are directly due to inappropriate experimental protocols that ignore the circuitry and complex functional organization of the human brain. This is a critical aspect that must be considered before drawing general conclusions and when comparing results across experimental studies. In particular, a specific attention should be paid to the control condition from which MI-related brain activity is reported. In fMRI data analyses, images of activity from experimental and control conditions are subtracted to determine significant peaks of activation. Hence, a difference which does not reach the significant threshold does not necessarily reflect a lack of activity, but can also be due to similar brain activations during both experimental and control tasks. In other words, if the control task is not thoroughly determined, the remaining activity of some brain regions might bias the pattern of activity observed during the experimental condition, i.e. MI. Inspection of the MI literature reveals great differences across studies. Among others, participants were asked either to listen for auditory stimuli without engaging in MI (perceptual control condition), having rest, projecting into a restful state, fixating a static stimulus on a screen, or even imagining a 'neutral' static motor task. Obviously, such differences might significantly influence the pattern of activity recorded when contrasting the experimental and the control conditions. This is likely to render the functional brain map, hence hindering the conclusions that one can reach. Another example is the fact that MI can be performed either with open or closed eyes. Marx et al. (2003) suggested that eye closure is likely to improve imagination, but in some circumstances, MI paradigms might require participants to open their eyes to fixate a screen and/or wait for instructions. As some activation or deactivation of cortical regions could be undetected with eyes open, however, MI should ideally be performed with closed eyes as it was asked for the rest condition (Marx et al., 2004).

Neural correlates of motor imagery
MI research provided convincing evidence that there is a functional equivalence between MI and motor performance (e.g., Jeannerod, 1994 . Taken together, these data support that MI and motor performance share the same neural substrate ( Figure 2), but interestingly there are also some differences within the pattern of activity in these areas, as well as with regards to the characteristics of the participants or the content of the MI task.

Contribution of the primary motor cortex
Since the pioneering PET study by Roland et al. (1980), who did not find a significant activation of the contralateral primary motor cortex (cM1) during MI, the question of the contribution of cM1 during imagined actions has attracted considerable attention and remains controversial (for reviews, see With regards to cM1 activation, Sharma et al. (2006) suggested that discrepancies in results may be due, at least partially, to methodological differences and difficulties in monitoring compliance with MI instructions. Definitely, the advent of fMRI techniques to investigate the contribution of cM1 during MI brought significant data into the debate. Furthermore, Sharma and colleagues argued that previous experimental studies did not specifically address whether the subdivisions of cM1 were differentially involved during MI. Furthermore, the primary motor cortex can be subdivided into an anterior component, thought to be executive in nature, while its posterior part would be involved in cognitive tasks or non-executive functions (Sanes & Donoghue, 2000). Accordingly, Sharma et al. (2008) reported that the cluster distribution in the anterior part of cM1 was significantly reduced during MI compared to the physical execution, while that of the posterior part was similar. Accordingly, they proposed that the role of cM1 and its subdivisions may be related to spatial encoding. Despite these results, determining how MI participates to the activation of cM1 remains difficult. Typically, cM1 activation during MI is usually smaller compared to that during overt motor execution, and is not systematically observed in all participants . It was also demonstrated that cM1 was increasingly involved along with the complexity of movements (Kuhtz-Buschbeck et al., 2003). As well, activation of cM1 might be differentially influenced by MI instructions, MI ability and motor expertise (Lotze & Zentgraf, 2010).
Taken together, and despite some controversial data, fMRI studies support that cM1 is actually activated during MI but more weakly than during actual movement. Such activation is not essential for imagery and the neurons are not in the same location than those active during movement execution (Lotze & Halsband, 2006

Activation of the other motor-related regions
Unlike the contribution of cM1, there is a general consensus regarding the activation of secondary motor areas of the cerebral cortex during MI. Accordingly, fMRI studies provided strong evidence that motor-related areas such as the ventral and dorsal parts of the premotor cortex (PMC), the supplementary motor area (SMA) and the pre-SMA, are active during MI of both simple and complex movements as well as subcortical areas such as the cerebellum and the putamen. Overall, it was shown that MI might activate a subset of the areas required for movement execution, albeit to a lesser extent (e.g., Macuga & Frey, 2011). Conversely, other activations could also be more robust in MI than during motor execution (e.g., Gerardin et al., 2000) thus leading to a partial overlapping of the networks mediating MI and actual execution of the same movement.  2008) further demonstrated that the SMA may inhibit activity of cM1 to prevent movement execution during MI. Interestingly, some authors also reported that the overlap of SMA activity during MI and motor execution is partial, and that the pre-SMA might rather play a specific role during imagined movements. Hence, the pattern of activity in this region could be higher during MI than during actual practice of the same task (Gerardin et al., 2000). Pre-SMA is known as being involved in acquiring new sequences, planning spatio-temporal actions, and updating motor plans for subsequent temporally ordered movements. As suggested by Malouin et al. (2003), the pre-SMA might thus be activated during MI to provide proper movement sequencing and timing. Taken together, MI would activate both the SMA and pre-SMA, not only for the preparation of the movement, but also for its suppressive influence on cM1 thus preventing movement execution.
Overlapping activity during MI and motor execution was also found in the premotor cortex . Although functional interactions between these anatomical systems are essential at the beginning of the learning process, there is ample evidence that the cerebellum plays a weaker role when the sequence is well learned and has reached asymptotic performance (for review, see Doyon & Benali, 2005), while basal ganglia remain active during the later stages of motor learning. By comparing the pattern of cerebral activations in 13 skilled and 15 unskilled imagers during both physical execution and MI sequence of finger movements,  showed that poor imagers not only need to recruit the cortico-striatal system, but also activate the cortico-cerebellar system as well. Findings also suggested that compared to poor imagers, good imagers would have a more efficient recruitment of movement engrams. Although this remains a working hypothesis awaiting experimental investigation, the authors concluded that the pattern of cerebral activation recorded during MI in poor imagers might improve and evolve close to that observed in good imagers after MI training. This might suggest that the expected changes in subcortical brain activations during MI would reflect those elicited by the process of actual motor learning (see also Lafleur et al., 2002;Jackson et al., 2003).

Activation of parietal regions
Activation of parietal areas including possibly the somatosensory cortex, but more certainly the inferior and superior parietal lobules and the precuneus, is frequently reported during MI ( Interestingly, bilateral parietal lesions resulted in a complete unawareness of movement execution during imagery (Schwoebel et al., 2002). In fact, the patient exhibited hand movements during the MI of the same body segments, but explicitly denied that they occurred. These data strongly support that parietal areas are critically involved in the generation and guidance of imagined movements. More importantly, they further suggest that the patient failed to inhibit the motor consequences of MI, and therefore, that parietal areas may also play a critical role in the inhibitory processes of the motor command during MI. Finally, and in conjunction with previous fMRI data (Gerardin et al., 2000), these findings suggest a functional dissociation between cortical areas underlying MI and motor execution. The primary sensory area would be more activated during motor execution, due to sensory feedback processes, whereas MI might elicit greater activity both in the inferior and superior parietal cortices, especially in the left hemisphere.

The effect of motor imagery content
A limited number of fMRI studies examined the influence of MI content on brain activations, and most specifically whether visual imagery and kinaesthetic imagery recruit different neural networks. In a first study looking at this issue, Binkofski et al. (2000) reported that the anterior part of the intraparietal sulcus was more active during kinaesthetic imagery of finger movements, while visual imagery yielded stronger activation in the posterior part. Bilateral activations were also reported during kinaesthetic imagery in the opercular portion of the ventral PMC. In a seminal paper, Solodkin et al. (2004) later investigated the neural networks mediating physical execution, visual imagery and kinaesthetic imagery of hand movements. Although these two types of imagery shared similar neural substrates, including the connection from the superior parietal lobule to the SMA, the main difference was found in the inputs from the superior parietal lobule and the SMA to cM1, which were opposite to those observed during motor execution. In the same line of research, Guillot et al. (2009) examined whether the same group of participants with high imagery abilities recruited similar or distinct brain activations during visual imagery and kinaesthetic imagery of complex hand movements. Visual imagery activated predominantly the visual pathways including the occipital regions and the precuneus, whereas the pattern of kinaesthetic imagery involved mainly motorassociated structures and the inferior parietal lobule (Figure 3). These data support the hypothesis that visual imagery of a motor sequence refers to the visual properties of visual perception, while kinaesthetic imagery includes in greater extent motor simulation processes related to the form and timing of actual movements (Michelon et al. 2006). Although visual and kinaesthetic imagery shared common brain structures, these data provide strong evidence that the patterns of neural activity mediating the ability to perform a specific MI type are partially different. Practically, it might suggest that participants are able to favor one sensory modality to form mental images, although MI is a multimodal and multidimensional construct. Similar data have been reported in regards to the visual imagery perspectives, that is, by comparing first and third person imagery perspectives. The first study looking at this issue, was conducted by Ruby and Decety (2001) with PET scan. They investigated the neural networks mediating MI when the participants were asked either to mentally simulate an action or to imagine someone else (i.e., the experimenter) performing the same movement. They found that the inferior parietal lobule and the somatosensory cortex were more activated during the first-person than the third-person perspective, while increased activity was observed in the precuneus and during the third-person perspective. Then, fMRI experiments were designed to examine in greater details whether using either the first or third-person imagery perspective lead to motor-areas-related changes. Jackson et al. (2006) explored the neural circuits involved in imitation and perspective-taking, and showed that first-person perspective was more tightly coupled to the sensory-motor system than the third-person perspective, which requires additional visuospatial transformation. More recently, Lorey et al.
(2009) also found stronger activation of motor and motor-related brain structures, like the inferior parietal lobe, when imagining a movement through a first-person compared to a thirdperson perspective. They argued that proprioceptive information on actual body posture would be more relevant for internal visual imagery users, which also reveals that bodily information is mainly integrated into the image of one's own body movement.
More generally, these results provide clear evidence that both imagery types (e.g. visual vs. kinaesthetic imagery) and perspectives (first-person vs. third-person imagery) are partially mediated through separate neural systems, which may therefore contribute differently during the process of motor learning and neurological rehabilitation. This assumption is supported by a remarkable fMRI study demonstrating that MI engages systematically the organized sections of cM1 in a somatotopic manner, i.e., that the content of the mental images is reflected in the pattern of motor cortical activation (Ehrsson et al., 2003).

Neural mechanisms for motor expertise
The fact that MI efficacy depends on the individual ability to form accurate mental images is now well-established. Although not systematic, this capacity to mentally simulate forthcoming actions may be influenced by the individual level of expertise in the corresponding motor task. While the concepts of motor expertise and MI ability measurements have been considered early on, researchers are able, for a short time, to assess the content of MI objectively using thorough procedures, in particular since the advent of functional brain mapping studies (Guillot et al., 2010). Taken overall, fMRI data strongly support the existence of distinct neural mechanisms of expertise in MI, as a function of the individual skill level. For instance, lower cortical activation was recorded in professional violinists as compared to amateurs (Lotze et al., 2003). Similar differences were observed in the neural networks mediating MI in novice and expert athletes (Milton et al., 2008). By comparing brain activations of six golfers of various handicaps during MI of a golf swing, Ross et al. (2003) found decreased activations of the SMA and cerebellum as a function of golf skill level, i.e. an inverse relationship between brain activity and skill level. Also, golf swing MI yielded few peaks of activation in the basal ganglia and cingulate gyri across all skill levels. Milton et al. (2007) finally reported that the posterior cingulate, the amygdala-forebrain complex, and the basal ganglia were activated in novices but not in elite golfers during motor planning of a golf swing movement, hence confirming that the neural networks controlling both motor planning and MI are influenced by the individual skill level. As a whole, changes in cerebral activations confirmed previous investigations showing that levels of expertise during the motor learning process are supported by different neural networks (Doyon & Benali, 2005).
Analogously, researchers investigated the neural networks mediating musical experience, training onset, and training stages. Langheim et al. (2002) first showed in participants who imagined playing a musical selection with their instrument that an associative network including the superior parietal lobule, the inferior frontal gyrus and the bilateral lateral cerebellum was activated during. This network would be particularly activated during the coordination of the complex spatial and timing components of musical performance. As mentioned above, Lotze et al. (2003) compared the patterns of brain activation during auditory imagery in experienced and novice musicians, with the professionals reporting frequent use of imagery with high vividness. Interestingly, experienced musicians recruited very few cerebral areas, while amateurs manifested a widely distributed activation map. In the professional group, however, more activation was observed in the SMA, the PMC, the superior parietal lobule and the cerebellum. Finally, Kleber et al. (2007) reported an increased activation in the fronto-parietal regions during imagined singing, suggesting increased involvement of working memory processes during imagery To summarize, fMRI data provided strong evidence that partially separate neural networks are activated during MI of both motor and musical performance in regards to the individual differences in the level of expertise. These findings therefore strongly support the hypothesis of distinct neural mechanisms for expertise in imagery, independently of the imagery type, with the network integrating the superior parietal lobule being seen as mediating the imagery activity of highly experienced people.
Apart from the level of expertise itself, the rate at which the movement is performed and imagined is also likely to recruit distinct neural patterns. Accordingly, Sauvage et al. (2013) nicely demonstrated that although execution and imagination of slow and fast movements activate a common neural substrate, the rate of movement differentially yielded associative cortical, striatal and cerebellar areas. Their data suggest that slow movement and motor imagery of both slow and fast movements activated fronto-polar, orbitofrontal and dorsolateral prefrontal cortices, while fast movements recruited more intensively the premotor cortex and the cerebellum.

fMRI advances in the study of motor imagery
Going beyond the usual comparison of brain activation patterns that are associated with MI and motor execution of the same movement is the next step in this field of research. To do so, researchers may take advantage of the recent methodological fMRI developments. Both the use of real time fMRI and the recording of changes in functional and effective connectivity in the neural networks activated during MI are of particular interest. Also, reporting more systematically negative BOLD responses might contribute to expand our knowledge on neuronal inhibition during MI.
It is now possible to image human brain functioning in real time with fMRI (Esposito et al., 2003;Weiskopf et al., 2003; for an extensive review, see deCharms, 2008). This technique provides a reconstruction of the raw data obtained with the brain scan, while the scan is going on. There are several exciting research perspectives for considering the contribution of real time fMRI dedicated to MI study. This technique may first be useful during the MI learning process, especially when the pattern of activation during mental simulation is not the one expected. Indeed, real time images are expected to provide participants some objective information related to the vividness of MI, since they have been educated to gain some familiarity with the neuroanatomy before neurofeedback fMRI sessions. Receiving a feedback from brain activation in predetermined regions of interest seems possible. An average 2-5s time lapse usually remains necessary, due to the physiological delay of the haemodynamic response (a more simple feedback can also be used, e.g. a simple score using a Likert-type scale). A well-known study supporting the use of real time fMRI was published by Yoo and Jolesz (2002), who used visual feedback of brain activation to guide participants to adjust their motor task performance and to achieve the desired modulation of cortical activity. For example, during simple hand movements, participants spontaneously involved more muscle groups and increased tapping frequency along fMRI sessions. The biofeedback given to participants with regard to the activated neural network helped them to modulate their cortical activation. A significant illustration of the strength of this methodology in the field of MI was offered by deCharms et al. (2004) during imagery of a manual action task. In this study, participants received feedback about the activation level in the somatomotor cortex with a simple virtual reality interface. The results showed that they enhanced the fMRI level of activation driven by MI in the somatomotor cortex through the course of training. Moreover, the activation of this region after MI training was as robust as that recorded during actual practice. In a more recent study, Yoo et al. (2008) showed that real time fMRI might help individuals to learn how to increase region-specific cortical activity associated with a MI task. Practically, the level of increased activation in motor areas was consolidated after the 2-week self-practice period. Regarding the method for presentation of neurofeedback (intermittent presentation or continuous presentation) and the nature of the neurofeedback being presented to participants during a MI task (true or false neurofeedback regarding brain activations), Johnson et al. (2010) further reported that the intermittent presentation of feedback was more effective than the continuous presentation in promoting self-modulation of brain activity. Accordingly, regular interruptions in neurofeedback presentation allowed central processing and integration of the information conveyed in the feedback regarding brain activations. The authors also reported that false feedback resulted in irrelevant brain activations with regards to the regions of interest targeted by experimental instructions. Finally, Xie et al. (2011) supported the effectiveness of delivering neurofeedback during MI using real time fMRI, and further provided evidence that the SMA was controllable by participants. These data strongly support that real time fMRI is a valuable technique to investigate whether participants are able to use a cognitive strategy to control a target brain region in real time. In the field of neurorehabilitation, for example, a similar approach could be used to learn how controlling pain by learning to control the brain regions that mediate pain perception (deCharms et al., 2005). Indeed, there are multiple therapeutic applications of real time fMRI but it is too early to predict success or failure and new experimental results are awaited.
Real time fMRI might also particularly useful in developing brain-computer interfaces (deCharms, 2008). A brain-computer interface is a novel communication system that translates human thoughts or intentions into a control signal without using any muscle activity (for review, see Pfurtscheller & Neuper, 2010). To date, many brain-computer interface systems have used MI tasks to modulate sensorimotor EEG activity taken to operate and control an external device. Pfurtscheller & Neuper (2010) nicely demonstrated how brain-computer interface systems using MI can contribute to help patients with various motor impairments and paralysis. Despite the temporal limitations of fMRI, using fMRI data for brain-computer interface remains plausible as many cognitive processes change slowly, over seconds or minutes (deCharms, 2008). Practically, the method is appealing although future experimental studies are necessary to determine its feasibility and effectiveness, such as during motor recovery of patients with motor impairment. Real time fMRI might also be used to explore the state of consciousness and communicate with patients in a persistent vegetative state. Accordingly, Owen et al. (2006) detected awareness in such a patient following instructions to mentally imagine moving around a house and playing tennis. Brain activations were observed in the parahippocampal gyrus, the posterior parietal lobe and the lateral PMC during MI of walking, and in the SMA when imagining playing tennis. These data provide evidence that real time fMRI might be used along with MI to actually communicate with people and/or patient who are physically or conventionally unable to interact with their environment (deCharms, 2008).
Creating functional connectivity maps of distinct spatial distributions of temporally correlated brain regions is another methodological advanced tool offered by fMRI. Functional and effective connectivity can be used to examine interactions among brain regions. Practically, these techniques go beyond the usual activation maps obtained through peak-detection methods, by looking respectively at temporal correlation between the time course of activation of two regions, and the influence of one neuronal population over another ( Furthermore, they found forward and backward effective connectivity between the SMA and the bilateral dorsal PMC, the contralateral primary and somatosensory cortex, and cM1. A last study by the same group of researchers confirmed these findings, and further showed that motor execution has some increased causal connections because of additional processes for the overt behavior stage (Gao et al., 2011). Taken together, these experimental studies highlight the advantages of studying functional and/or effective connectivity through fMRI to expand our understanding of the neural underpinnings of MI (see also Szameitat et al., 2012).
Reporting an elevation in the fMRI BOLD signal, namely positive BOLD, has become a common routine for mapping neural activity in the human brain. Aside from positive BOLD signal changes, several studies have also observed negative BOLD responses. The negative BOLD signal is a physiologic process correlated with a corresponding decrease in cerebral blood flow, oxygen consumption, and neuronal activity (for review, see Shmuel et al., 2002). Practically, negative BOLD signal reflects less neural processing for a given task as compared to a baseline condition. The negative BOLD response might be caused by a reduction in cerebral blood flow and is associated with decreases in the rate of oxygen consumption. Its neurobiological mechanisms are yet not well understood, and much of the debate has centered on whether its source is primarily vascular or neuronal. The neuronal origin might reflects a suppression of the local neuronal activity and/or a reduction in the afferent input, whereas the vascular origin refers to a reduction in cerebral blood flow to the less demanding regions due to the increase in flow to the demanding areas, without a necessary decrease in neuronal activity in the negative regions (Schmuel et al., 2002). Until recently, very few experimental studies were designed to analyze negative BOLD responses during either visual or MMI tasks. An interesting and innovative paper by Amedi et al. (2005) investigated the pattern of brain deactivation during visual imagery and compared it to the neural correlates of visual perception. While they found that visual imagery and visual perception share similar neural substrate, these two conditions yielded different brain-deactivation profiles as shown by negative BOLD responses. Of particular interest is that the authors reported a robust deactivation in early auditory areas as well as a selective deactivation in the somatosensory cortex during visual imagery, hence supporting that visual imagery is associated with deactivation of nonvisual sensory processing. Based on these data, they stated that that pure visual imagery might be characterized by an isolated activation of visual cortical areas with concurrent deactivation of sensory inputs that may potentially disrupt the image created by the mind's eye. They also considered that deactivation could be the consequence of filtering out irrelevant stimuli. The correlation between the level of deactivation and the vividness of visual imagery might support the hypothesis that participants with high imagery ability are able to shut down or disconnect the ''irrelevant'' cortices. To the best of our knowledge, unfortunately, there are no experimental studies specifically designed to look for similar results in MI paradigms per se. Interestingly, it has been suggested that the prolonged negative BOLD signal reflects an inhibition or suppression of neuronal firing patterns (Raichle, 1998;Kastrup et al., 2008). Looking more systematically for negative BOLD responses might therefore be of particular interest to investigate the inhibitory processes of the motor command during imagined movements. Indeed, motor command inhibition mechanisms during MI remains rather unclear since no direct fMRI brain signal is known to characterize neural inhibitory processes, and inhibition during MI has been primarily studied from brain activation signals. The negative BOLD response might therefore be an interesting methodological complement to extend our knowledge of motor inhibition

Conclusion
The data reviewed in this chapter strongly support that recording human brain functioning with fMRI during MI provides an objective measurement of the neural networks underlying MI processes. Although they are not totally overlapping, it is now well-established that the neural substrates mediating MI and actual execution of the same task are quite similar. Spurred by recent fMRI methodological advances, the next step of MI research will certainly contribute to understand in greater details the neural correlates of imagined movements, as well as the inhibition of the motor command. MI studies could therefore benefit from the use of real time fMRI, effective connectivity, and also negative BOLD. This latter technique might be of particular interest in investigating the inhibitory processes of the motor command during MI. Cerebral regions which are inhibited during mental operations or human behavior are certainly of the same scientific interest than those which are activated.