Session Summaries of "Neuroimaging in the Study of Neural Repair and Rehabilitation", June 1-2, 2007
1. Session Title: Clarifying Attentional Mechanisms and Studying Neurologic Attention Deficits (Using Imaging).
Facilitators: Ayelet Sapir, Branch Coslett
Summarizer: Patricia Arenth
Proposed Discussion Questions:
What experimental tasks and paradigms are most useful for understanding neural control of attention?
How well do imaging tasks capture ecologically important phenomena such as sustained performance, performance in unstructured settings, etc.?
Can modulation of sensory or motor systems be useful in measuring attentional phenomena?
What are the implications of individual and group differences in task difficulty, skill level, and effort, in understanding attentional systems?
Summary of Discussion:
Initially, facilitators discussed basic definitions/descriptions of attentional processes as follows:
Attention was described as the process through which the mind chooses among various the stimuli striking the senses at any given moment. It was pointed out that we are constantly bombarded with more stimuli than we are able to process, and our attentional systems filter out the most salient information using bottom-up or top-down processes. Brief descriptions of orienting attention, selective attention, divided attention and automaticity were given.
A couple of examples of attentional studies were provided:
In one description, subjects engaged in a visual task where a visually presented arrow provided an anticipatory cue for the upcoming target location. Reaction time was faster for attended vs. unattended targets, suggesting that spatial attention facilitates visual processing.
A second example provided was of dichotic listening tasks, in which individuals have been shown to attend to auditory information coming into one ear, while ignoring additional information coming into the other (i.e., the well known “cocktail party effect”).
The facilitators suggested that imaging techniques may be helpful in studying several different aspects of attention. For example, it may be helpful to use imaging to evaluate how individuals shift attention from one set of stimuli to another – i.e., is attention shifting done by enhancing a given target, or by suppressing distractors. Similarly, it may be helpful to use imaging to evaluate how we filter information, and where in the attentional process the filter is located. Imaging may contribute to evaluating how we select relevant targets, whether there is early or late selection, and may help us in evaluating between different types of attentional processes in use during various circumstances (i.e., endogenous vs. exogenous orienting, sustained attention vs. shifts in attention).
At this point in the presentation, there was a discussion among participants regarding at what point filtering occurs. Dr’s Seitz and Shomstein discussed early filtering. Dr. Whyte brought in more recent discussions about flexible placement of filters in accordance with task demands. Dr. Shomstein made a point regarding the salience of the stimulus and the interaction of top-down and bottom-up processes and suggested that the goal of attention may shift if more salient information interrupts current processing, and that this could be either a bottom-up or top-down process.
Dr. Coslett indicated that attention has an effect on the BOLD signal, and that imaging studies are warranted to look at this further: He asserted that attentional processes augment/operate on the area(s) of the cortex most relevant to the requirements of any task presented to a subject (i.e., visual, motor etc.). As a result, the relevant cortices will show changes in activation depending on the interaction of the task demands and attention. In looking at the BOLD signal, it is difficult to define what is due to attention and what is not.
Group discussion focused on how difficult it is to define what is attention and what is not. Dr. Detre asked a question about whether there are good ways to control for attentional modulation in sensorimotor tasks (i.e., vary the nature of stimulus parameters in the study) and whether these things should be controlled? (i.e., reduced performance in a person who has never had an fMRI and is distracted by the environment – should you have “practice” for all subjects to control). Dr. Shomstein suggested that it is not possible to get rid of attentional issues, but that by varying vs. keeping attentional demands the same across tasks, it might be possible to control them or use as a covariate. Dr. Detre asserted that a problem is that a measure of attention is always an inference, and that the individual subject’s capacity for attention is also always a factor.
Some additional studies were presented as examples of the effects of attention on performance:
Baker and Sands showed performance differences when subjects completed a finger-tapping task while looking at their finger versus looking away. Similarly, John Bradshaw showed that where one allocates visual attention there is a huge effect on performance.***References?
Dr. Selzer asserted that multiple mechanisms of inattention could be likely: For example, inattention could be due to habituation over time (it is possible that in some instances, individuals my habituate too quickly, reducing attention). In some instances (i.e., individuals with hemiplegia), failure of input may be at play. Fatigue may also be a factor.
Dr. Seitz indicated that even when encouraging normal subjects to attend to electrical stimulation, the threshold had to be increased over time, suggesting that peripheral mechanisms may contribute significantly.
A question about how to differentiate between habituation and inattention was raised. Dr. Cosett asserted that attention occurs at all levels and that habituation could be considered a form of physiological inattention.
Dr. Sapir suggested that one limitation of attentional studies is that in the lab, we attempt to control external stimuli, however in the real world there are more stimuli that we need to suppress. Dr. Shomstein reported an example of a video study that tries to replicate the real world situation in the lab: When subjects were asked to focus on and count the number of times individuals in the video passed a ball back and forth, only 35% of subjects noted that a person dressed in a gorilla suit came into the picture while the ball was being passed. ***reference?
It was suggested that the way in which we process visual and auditory information may be allocated to different spatial locations in healthy individuals, and that deficits in spatial orienting in individuals with mild TBI could account for some attention issues found in this group.
The Posner task of visual attention was discussed.
Discussions regarding spatial orienting using exogenous (reflexive/bottom-up) versus endogenous (top-down) processes followed.
Dr. Mayer presented a study where subjects were presented with identical tasks in both visual and auditory formats. By instructing subjects to selectively attend to the visual or auditory stimuli (i.e., finger tap to visual input vs. tap to auditory), it was possible to look at differentiated activation patterns using neuroimaging. ***reference? Others in the group suggested the possibility of using a numerical Stroop task, or using parametric manipulation of difficulty.
It was asserted that frontal/parietal areas are known to be involved in attention, working memory, and spatial attention. A question was asked about whether spatial working memory and spatial attention could be considered to be the same thing. It was suggested through the discussion that both utilize a similar scheme – holding something online to cognitively process information. Whether these processes are exactly the same was uncertain.
The discussion then returned to the question about attention versus habituation: It was suggested that habituation requires an initial presentation/attention to stimuli, and then habituation occurs, however inattention may not require the initial presentation. At this point, there was a discussion about how processes such as priming, gating, and other “unconscious” processes fit into the model of attention. It was decided that, from a patient standpoint, keeping the focus of discussion on the study of attentional processes that involve conscious awareness may be most beneficial for the purposes of this discussion.
The focus of the discussion moved to a discussion of patients with visual neglect, which usually occurs after a right hemispheric stroke. Neglect refers to the inattention to contralateral visual information. Additional symptoms of right hemispheric stroke often also include a lateral rightward bias, as well as deficits in arousal/vigilance and spatial working memory. A neuroimaging study was presented in which a group of 11 stroke patients were followed. The design of the study was as follows:***Reference?
Subjects were first seen at 3-4 weeks post stroke. At that time, they completed a neuropsych battery, a computerized battery (orienting, motor and vigilance tasks) and a Posner cueing paradigm in the fMRI scanner with an event related design. At approximately 39 weeks, these tasks were repeated and a high resolution anatomical scan was obtained.
Based on behavior– subjects recovered in terms of reaction time and ability to respond when an invalid target cue was presented.
Acute neglect: results suggested that the whole brain shuts down: neuroimaging showed a lesion on the right, with very little activation, suggesting reduced whole brain activation. High activity was found in the left parietal cortex of acute patients.
Chronic neglect: neuroimaging results suggested activity levels were returning, even on the lesioned side.
Acute vs. Chronic voxel wise ANOVA results: higher activity at chronic versus acute stage. In the left superior parietal lobule, acute patients actually had hyperactivity as compared to chronic patients. Activity on the Left SPL was correlated with neglect severity. The BOLD % signal change increased positively with more severe neglect in the chronic phase. There was not enough power in this study to see correlations at the acute phases (not able to predict neglect at the chronic phase from acute BOLD change).
A suggestion was made about using voxelwise ANOVA versus regions of interest for study of these types of issues.
Participants brought up several potentially relevant studies and models: Nets and Turtain paper, Herald Model, Cabezza Model (uncertain of proper spellings and years for each of these).***Check names and provide references?
Dr. Coslett mentioned a 2006 study by Smith, Taylor, Bramme, Toon et al. in which individuals with ADHD compared to healthy controls.***Reference? Participants participated in a “go-no go task as well as a motor Stroop task with congruent and incongruent cues as well as an oddball task in which they were asked to do nothing. They were also presented with a “switch task”. Behavioral data was comparable. Neuroimaging data was similar for the go-no go task and Stroop (which were considered to be matched tasks), but were significantly different with a more complex task (the “Switch Task”). Findings indicated that seemingly similar tasks could produce significantly different neuroimaging results.
Models and theories of attention were discussed, and behavioral and neuroimaging studies of attention were presented. Consistent with much of the discussion throughout the symposium, there was not always consensus on the theories and definitions of attention or on the best methods for studying the processes of attention. It was suggested that neuroimaging techniques may be helpful in differentiating between various attentional processes, however it was also asserted that attention is a difficult construct to define and study. It was asserted that control of attention is difficult to maintain in experimental design, and is often inferred, but difficult to measure as there is still much to learn in this area. Previous studies have indicated that attention does have an effect on task performance and on neuroimaging results. There is still much to be done in understanding the complexity of cognitive processes, as well as in understanding the reliability and results gained through our use of neuroimaging tools. Additional complexity is added by studying patient as well as healthy populations. Overall, the discussion seemed to indicate that there are many challenges, but also great potential for learning more about attentional processes through the use of neuroimaging techniques.
2. Session Title: Serial Neuroimaging Over Time
Facilitators: John Detre, Richard Wise
Summarizer: Sandy McCombe Waller
Proposed Discussion Questions:
1) For motor activation paradigms, what are the better techniques to employ: e.g., how should we interpret a change in TMS excitability, BOLD magnitude or volume, and ROI and remote distribution of activity over time after stroke?
2) Can we establish standards for reproducibility in longitudinal studies? How should we account for changes in performance, effort, and experience on an acutely evoked signal at the moment of testing given the ongoing neurobiological changes of the brain over weeks and months? What are the best movement (kinematics, force, directionality, speed, normalcy of the action, etc.) and statistical methods to guide the interpretation of these changes over time within and across subjects and between a single subject and a large control group?
Summary of the presentations:
To start the session off, John Detre presented some of the issues relating to the use of fMRI serially over time. Citing studies including Thulborn, 1999 , Karbe 1998 , Saur et al 2006  it was pointed out inconsistencies found with fMRI studies. In some studies activation shifted from contralesional to ipsilesional and correlated with functional outcome. Other showed contralesional activation did not predict a good functional outcome. Activation is not clearly understood as to what is really going on neurally in the brain. We can not be sure the location of activation necessarily represents areas of neural activity that relates to function. There are limitations in using inferences about neural function using BOLD in serial studies. It is dangerous to infer physiology from a p-value.
Richard Wise (facilitator) then offered a brief presentation of alternative uses for serial imaging providing examples from research of Speech and Rotated speech. One suggestion was the use of serial imaging not for prediction but to suggest treatment. It was recommended to take an established stroke syndrome - even if the patient problem is rare. Understand clearly the lesion and see what images are derived prior to and after intervention. Clever experimental design can help deal with minimizing the performance confound (particularly in language studies).
Summary of the Discussion:
Serial imaging could be useful in predicting outcome of the patient and possibly in elucidating neural mechanisms of recovery. Much of the discussion focused on what result in an fMRI study actually provides insight into mechanism. Proposed suggestions included: 1) Deviations from a previously defined “normal” network or pattern of activation, 2) A pattern seen at time point one that predicts subsequent recovery or lack thereof at time point two, 3) Identification of critical periods/ windows for therapeutic interventions based on the evolution of activation patterns. However there remains a concern that fMRI primarily reflects task execution, and the neural substrate for execution of the task and for recovery may be two different things.
Serial imaging could also be useful in characterizing the time course of response to an intervention. The experimental paradigm should be as constrained as possible to limit the performance confound. It may be possible to control for performance effects by comparing delta versus absolute measures, thereby controlling for baseline performance. An alternative strategy is to examine resting (low frequency) BOLD connectivity since this may be independent of a performance confound. However some discussants felt that resting BOLD only relates to resting function and if you want to see what the brain is doing during activation, you need to measure during activation.
When using fMRI in a test/ retest manner in order to look at the impact of an intervention one should keep mind of the following: 1) Pick an intervention that has positive functional or impairment outcomes, 2) have a hypothesis/ theory as to what you think the intervention is doing to guide where you will look with fMRI. This will help in being able to related the changes seen in fMRI to the intervention versus to just a change in behavior.
Interventions were discussed at length without a clear consensus on what interventions are best to use. Some argued for interventions that make changes on an impairment level (not generalizable necessarily to function performance), other felt that interventions needed to show meaningful change in function. One point of consensus was that interventions should be:
a) theory driven (one should have a clear idea of what one thinks the intervention is doing and to use that information to guide applications of serial imaging
b) shown to be effective in bringing about a change in behavior (proof of principle)
c) studied with respect to the activation patterns associated with being a treatment responder/ non-responder
When should you do serial imaging / how do you decide which time points you will use?
One typically chooses to use pre/ post intervention or scanning after you see some sort of behavioral change/ improvement. Multiple scan may be useful in tracking time-course of recovery as well as using them as serial predictors.
What are the reliable biomarkers?
Spatial pattern effects may be better markers for change than just use of magnitude changes with serial imaging.
A interesting result would be one in which a behavior is regained after injury AND the activation is different from “normal”. This is an interesting finding that might suggest a recovery mechanism that can be studied in more detail.
What are the best approaches (both for fMRI and performance outcome measures)?
A) One suggestion is the use of ASL (perfusion fMRI). This technique can allow one to evaluate adequacy of baseline perfusion to determine if one can even get a BOLD signal. It can be helpful in reducing the noise (low frequency drift) in fMRI. ASL may be particularly useful when testing using low task frequency (e.g. long task blocks, repeated scanning over days/weeks). Other benefits of ASL include that it measures a biological parameter exclusive of the subject and is stable to hardware changes which makes it’s use conducive for multisite trials.
B) The use of DTI was also proposed. It may be better to use structural measures (that are independent of variation seen with functional activation measures). Is it possible that DTI would provide a more stable measure to relate to functional change?
C) Possibly combinations of methods will provide the greatest insight (PET and fMRI, EEG and fMRI, TMS and fMRI for examples)
D) Consistent use of performance outcome measures across studies would be helpful. Measures vary or are not fully explained. Some measures are purely impairment based measures while others are functional performance measures.
E) What are the best analysis methods? Are there new ways of evaluating data that improve interpretability of fMRI findings? New analyses beyond just pre/ post analyses were suggested and discussed in more detail in the previous session.
F) Single case designs is another approach – comparing the various patterns of activation seen with single cases. This could be done statistically if patterns are determined to be different.
Serial imaging has its limitations but is still valuable if care is taken in experimental design, data collection and data analyses. Recommendations relating to type of imaging to use include Perfusion fMRI and use of DTI. These are newer approaches that address some of the limitations of using BOLD fMRI. These have their limitations as well and should be considered to avoid over-interpretation of data. Combinations of methods such as PET and fMRI, EEG and fMRI, TMS and fMRI may provide the best insight. When using serial imaging it is important to try to control for the performance confound. One suggestion was to make sure task completion in scanner was something the patient could do at all time points and that it was tightly controlled. It is important to consider the functional outcomes that are used in imaging studies as well as the interventions that may be used to understand mechanisms associated with recovery. Outcomes should be standardized and ideally include not only impairment but functional performance measures as well. Interventions used in serial imaging studies should be know to bring about functional performance gains, should be theory driven and with a hypothesis as to how the intervention is thought to provoke the system. Targeted analyses designed a priori can then be carried out, to shed light on mechanisms of recovery. Different experimental designs are suggested in the use of serial imaging going beyond pre/ post design studies. Examples include single case designs to characterize recovery patterns in detail, or studies of patients with well established pathology, comparisons of patterns of activation in responders and nonresponders with a particular interest in studying those with positive behavioral outcomes and new patterns of activation.
Thulborn, K.R., P.A. Carpenter and M.A. Just, Plasticity of language-related brain function during recovery from stroke. Stroke, 1999. 30(4): p. 749-54.
Karbe, H., A. Thiel, G. Weber-Luxenburger, K. Herholz, J. Kessler and W.D. Heiss, Brain plasticity in poststroke aphasia: what is the contribution of the right hemisphere? Brain Lang, 1998. 64(2): p. 215-30.
Saur, D., R. Lange, A. Baumgaertner, V. Schraknepper, K. Willmes, M. Rijntjes and C. Weiller, Dynamics of language reorganization after stroke. Brain, 2006. 129(Pt 6): p. 1371-84.
3. Session Title: Neuroimaging as a physiological marker to guide a strategy for rehabilitation
Facilitators: Steve Cramer, Bruce Dobkin
Summarizer: Carmen Cirstea
Proposed Discussion Questions:
Can a therapy for aphasia, paresis, neglect etc. be chosen based on how it alters a short-term TMS, fMRI or the response to other techniques, i.e., the intervention appears to engage or fails to alter the expected regions of interest?
Can these techniques help define the optimal intensity and duration of a therapy using repeated measures over the time of treatment?
What intervention, if any, should serve as the control condition for a training paradigm plus neuroimaging study executed over weeks or months?
What set of longitudinal data are needed to make reliable brain-behavior correlations as training proceeds?
The advantages of using uniform clinical outcome measures between studies.
Can functional neuroimaging offer information that is not clinically evident?
Summary of Discussion:
While there is increasing evidence that the neuroimaging techniques might be used in stroke studies, less is known about their relationship with functional outcome following stroke. More precisely, how do these new neuroimaging modalities compare with clinical testing) in assessing the potential of patients to respond to treatments. However, neuroimaging studies of stroke survivors are not every day clinical practice while the clinical scales are less expensive and more practical. Since the time since stroke and current clinical scores seem not to be good predictors of the potential for further recovery in late stroke, their combination with neuroimaging data may allow us to gain predictive power. In addition, neuroimaging biomarkers may increase test sensitivity by being less exposed to subjective interpretation compared to the clinical biomarkers. Furthermore, these biomarkers would be especially useful for the design of randomized clinical trials to estimate sample size and define inclusion criteria, i.e., including only those patients with potential for treatment response.
The facilitators suggested the goal of developing predictors of the capacity of patients to make further improvements in their sensorimotor and cognitive impairments, and consequently to identify the patients who are unlikely to make a clinically meaningful neurological recovery within the context of a specified strategy for acute stroke therapy, rehabilitation or neural repair.
This approach could justify withdrawal from a rehabilitation intervention and administration of other possible interventions to these patients. It is clear that an ideal prognostic biomarker should be readily available, easily reproducible, and associated with a high degree of specificity for poor outcome.
This session concentrated on the role of neuroimaging techniques, in particular fMRI, as a potential predictor of further functional improvement. For example, Dr. Cramer has shown that in 24 stroke patients who participated in 6 weeks of rehabilitation therapy with or without motor cortex stimulation, the best predictor for treatment gains was lower motor cortex BOLD activation 1Another study mentioned during this section was that of Dong et al., 2006 about the role of serial fMRI as a physiological indicator for “dose-response” interactions during a task-specific intervention. More precisely, by using fMRI, the M1 activation in 8 moderately impaired stroke patients was measured midway through a 2-week arm-focused intervention in order to capture adaptations induced by the initial week of training. These midway M1 adaptations (ipsilateral hemisphere) were used to anticipate post-therapeutic behavioral changes in impaired hand function. Furthermore, this correlation between brain activation adaptations and behavioral improvement, called “brain-behavior correspondence” by the authors combined with initial impairment level might be used to guide the optimal duration for this task-specific therapy. Similar data were provided by Koski et al2, whose data suggested that TMS measures of the corticomotor pathways across one day of therapy might predict the treatment-induced plasticity and behavioral gains (for upper extremity) over subsequent weeks in patients with moderate to more impaired deficits after stroke.
However, when using fMRI to study neural response in patients with stroke, there are several issues to consider. Although detailed theoretical background to this technique was beyond the scope of this session, the facilitators pointed out the impact of impaired cerebrovascular reserve or advanced narrowing of the cerebral arteries on the BOLD signal (see Hamzei et al., 2003; Rossini et al., 2004)3, 4. Additionally, it still is not clear how the BOLD signal is affected by parameters such as time after stroke and large or small vessel disease. Consequently, a multimodality approach using different imaging techniques (BOLD, perfusion scanning) and concurrent neurophysiological methods (EEG, MEG, TMS) was proposed to address the influence of different physiologic variables. For example, Stinear et al (2007)5 by using a multimodality approach (TMS, structural and functional MRI) characterized the state of the motor system in chronic stroke patients to predict the functional gains made in a subsequent motor practice intervention. The structural integrity of the CST was assessed by the presence or absence of MEPs (TMS) in the affected arm, and FA values of PLIC (DTI), while the lateralization of brain activation during a motor task was assessed by fMRI. The TMS measures suggested that in patients with MEPs, meaningful motor gains were still possible 3 years after stroke while in patients without MEPs, limited gains were reported. In contrast, the fMRI measures (i.e., brain activity lateralization) were not able to predict future functional gains through motor practice. However, that might be explained by the motor task evaluated during fMRI (self paced opening and closing of the hand that might induce variability in performance).
The next key question addressed in this session was the potential to study the brain’s capacity to be changed by a perturbation such as TMS to temporally and briefly manipulate cortical physiology. Numerous examples exist in the literature whereby a virtual lesion created by TMS provided insight into how function of specific brain regions related to behavioral status. This approach may serve as a behaviorally independent assay of connectivity between cortical areas.
Another important question was whether these changes in motor system reorganization might help define the optimal intensity and duration of a therapy. Therapy can be considered as an input that interacts with a damaged system (in our case, stroke) and the aim of this input is to optimize the functional reorganization of this system. This input may succeed in driving functionally useful changes if it interacts with intact brain regions and networks that influence motor output pathways. Pharmacological and repetitive TMS therapies, may also drive activity-dependent spared pathways. Genetic factors could be a source of important interactions. For example, in the study of Kleim et al. (2006)6, the presence of the BDNF polymorphism was found to modify experience-dependent plasticity in corticospinal output in healthy subjects.
Whether changes in fMRI patterns predict skill recovery was another important question. The answer may depend on patient criteria selection, including only well recovered patients who could perform a distal motor task at baseline and patients with a small lesion affecting only M1 or subcortical white matter. \Ward et al (2003, 2004)7, 8 studied motor system changes by using fMRI in a such a group of patients, suggesting that during recovery from stroke, the nervous system may retain the ability to exploit the redundancy of the motor system by using intact and connected motor regions (i.e., secondary motor areas in the impaired and unimpaired hemispheres) to generate motor output. In other words, damage of a motor network could be partially compensated by activity in another region. However, the functional relevance of recruitment of these secondary motor regions is not fully understood. Based on the previous studies and following the Symposium discussion, two approaches were suggested to better understand neural substrates of functional reorganization: (i) use TMS to transitively disrupt these regions and measure differential behavioral effects in stroke patients vs. healthy, an approach that in a handful of studies does support that increased activity in secondary brain areas contributes to the final behavioral state (eg Lotze et al, J Neurosci, 2006)9; and (ii) measure how task-related activity covaries with modulation of task parameters (i.e., force modulation). Less is known about the functional changes underlying therapy-induced recovery in more severely impaired patients.
However, it is unlikely that the cerebral motor system response to injury involves a simple substitution of one motor region with another by taking new roles. Reorganization is often not successful in returning motor function to normal and seems to be determined by the extent of anatomic damage (damage of cortical motor regions, white matter pathways, and even which hemisphere is affected).
Cross sectional versus longitudinal studies
There are fewer longitudinal fMRI studies compared to cross sectional studies in stroke patients. The longitudinal studies generally suggest that functional recovery is associated with a lateralization of the task-related brain activation patterns towards the ipsilesional hemisphere, ie contralateral to movement. However, it is impossible to know whether the brain activation patterns returned to pre-stroke levels given persistent structural damage. From cross-sectional studies, it is clear that brain activation patterns will not return to normal in most of the cases. Thus, the longitudinal changes of motor activation patterns may represents an increase in efficiency, synaptic efficacy, and learning within the spared motor networks (Dobkin, Neurorehabil Neural Repair, 2007). The focusing of task-related brain activation will tend toward the most efficient system available. It is clear that brain activation pattern of an individual patient at one time point represents the state of reorganization within the available system and depends on a number of factors: anatomical lesion, premorbid state of the brain, drug treatments, genetic status (see previous section). All these factors influence the potential activity-driven changes within the intact motor networks, the putative mechanism of motor therapy.
The advantage of uniform clinical outcome measures between studies.
The facilitators pointed out the variability of clinical outcome measures between studies using functional neuroimaging techniques, in particular fMRI for understanding treatment effects after stroke.
However, it is clear that stroke-related impairments (i.e., motor, sensory, cognitive, language, etc) recover to different extents at different rates after stroke and the stroke treatment may affect various stroke-related impairments. Most studies in stroke treatments rely on composite clinical rating scales as outcome measures (i.e., modified Rankin scale, NIHSS). By using these composite scales, an important gain in one domain might be over or under estimated as little or no gain in another, when the domains are unequally weighted (i.e., motor vs. cognitive). The facilitators suggested using specific scales as primary outcome measure that address only one domain that is relevant to the rehabilitation intervention (i.e., motor recovery). For example, Dr. Cramer suggested using the Fugl-Meyer scale to assess upper extremity impairment. In addition, by using these specific scales, a stroke recovery treatment might be described as a treatment to improve a specific neurological function, such as “recovery of arm motor function”. Furthermore, the best primary outcome measure seems to be the within-subject change scores (for example, a change in FM score of 4 may be clinically significant) rather than a single cross-sectional value. Consequently, by using within-subject change scores, the sample size requirements of clinical trial may be reduced. Use of uniform measures across studies has been widely adopted, with success in dementia, multiple sclerosis, and spinal cord injury, and is much needed in the field of stroke recovery.
Can functional neuroimaging offer information that is not clinically evident?
The overall opinion was yes. Functional neuroimaging provides insights into biological mechanisms underlying functional recovery and potential targets for different therapies when anatomic imaging or behavioral assessment do not (i.e. Yozbatiran et al., 2006)10. In other words, the value of functional neuroimaging is to predict treatment response over time of an intervention and possibly to triage patients according to their brain’s physiological status rather than only by the clinical examination.
During the Symposium, the focus of the discussion was on functional neuroimaging biomarkers at the level of neuronal systems, rather than cells or molecules. Both approaches (systems and cellular) have something to teach each other about brain plasticity following injury. Magnetic resonance spectroscopy (1H-MRS) provides an imaging probe of neuronal metabolism (cellular level) that measures brain metabolites including N-acetylaspartate (NAA), a compound localized exclusively in neurons and their dendritic and axonal processes. A decrease of NAA concentration has been described in the brain lesion’s core, indicating neuronal loss, and in remote brain regions that appear normal on conventional MRI, indicating decreased neuronal metabolism. However, there is currently no direct evidence that changes at the cellular (i.e., decreased NAA concentrations) level are related to changes at the systemic (i.e., increased/decreased motor-related activation) level. Furthermore, the relationship between these cellular changes and functional recovery is unknown.
Should therapeutic resources be focused on patients with good prognosis or should increased resources be channeled to patients with poor prognosis?
Changes in motor performance over time poses a neuroimaging confounder. A task is needed that probes the changeability of the cerebral motor system in the context of motor practice but how the task is performed over time is understood in terms of changes in electromyographic, kinematic, and spatiotemporal measures that can be monitored during scanning to allow an accurate interpretation of neuroimaging data.
Multiple sites are needed to enter enough subjects into an fMRI-rehabilitation trial after stroke or TBI. Data collected across sites ought to be more generalizable than the usual single site trial that includes fewer than 10 subjects. However, issues in standardizing the imaging techniques across centers are not fully resolved.
1. Cramer SC, Parrish TB, Levy RM, Stebbins GT, Ruland SD, Lowry DW, Trouard TP, Squire SW, Weinand ME, Savage CR, Wilkinson SB, Juranek J, Leu SY, Himes DM. Predicting functional gains in a stroke trial. Stroke. 2007;38:2108-2114
2. Koski L, Mernar T, Dobkin B. Immediate and long-term changes in corticomotor output in response to rehabilitation: Correlation with functional improvements in chronic stroke. Neurorehabil Neural Repair. 2004;18:230-249
3. Rossini P, Altamura C, Ferretti A, Vernieri F, Zappasodi F, Caulo M, Pizzella V, Del Gratta C, Romani G, Tecchio F. Does cerebrovascular disease affect the coupling between neuronal activity and local haemodynamics? Brain. 2004;127:99-110
4. Hamzei F, Knab R, Weiller C, Rother J. The influence of extra- and intracranial artery disease on the bold signal in fmri. Neuroimage. 2003;20:1393-1399
5. Stinear CM, Barber PA, Smale PR, Coxon JP, Fleming MK, Byblow WD. Functional potential in chronic stroke patients depends on corticospinal tract integrity. Brain. 2007;130:170-180
6. Kleim JA, Chan S, Pringle E, Schallert K, Procaccio V, Jimenez R, Cramer SC. Bdnf val66met polymorphism is associated with modified experience-dependent plasticity in human motor cortex. Nat Neurosci. 2006;9:735-737
7. Ward N, Brown M, Thompson A, Frackowiak R. Neural correlates of outcome after stroke: A cross-sectional fmri study. Brain. 2003;126:1430-1448
8. Ward N, Brown M, Thompson A, Frackowiak R. The influence of time after stroke on brain activations during a motor task. Ann Neurol. 2004;55:829-834
9. Lotze M, Markert J, Sauseng P, Hoppe J, Plewnia C, Gerloff C. The role of multiple contralesional motor areas for complex hand movements after internal capsular lesion. J Neurosci. 2006;26:6096-6102
10. Yozbatiran N, Cramer SC. Imaging motor recovery after stroke. NeuroRx. 2006;3:482-488
11. Dobkin B. Confounders in rehabilitation trials of task-oriented training: lessons from the designs
of the EXCITE and SCILT multicenter trials. .Neurorehabil Neural Repair. 2007;21(1):3-13.
4. Session Title: Functional neuroimaging within hours and days after stroke or brain injury
Facilitators: Argye Hillis, Randolph Marshall)
Summarizer: Jacquie Kurland)
Questions for Discussion:
What limitations and opportunities are posed by neural, metabolic, and blood flow/volume factors?
Can fMRI, TMS, NIRS, etc complement the assessment of diffusion-perfusion MR imaging, DTI or spectroscopy for viable tissue that may be available to subserve later recovery?
What would be a set of standard activation paradigms to test for this possibility?
Limitations & opportunities re: blood flow, etc. Hemodynamic Changes in Subacute and Acute Poststroke Stages
It is important to consider issues related tohemodynamic changes in the hours and days after stroke or brain injury. Acute stroke is an evolving event in which there may be opportunities to intervene over several days following the insult. Dysfunctional regions within the ischemic penumbra may include viable tissue, but how long it will survive without intervention is unknown.
Hypoperfusion and Reperfusion
Restoration of blood flow is an important mechanism supporting recovery of function in the acute stage. This mechanism can be studied pre- and post-intervention using MR perfusion weighted imaging (PWI) to assess severity of hypoperfusion, diffusion weighted imaging (DWI) to assess extent of lesion, and language or other functional testing. Reperfusion of hypoperfused regions that correlate with improvements in language performance in acute aphasia patients can also provide evidence of regions that are essential to linguistic functions (Hillis et al., 2006). For example, Hillis et al. (2001) performed detailed single subject studies on six patients who had small infarcts surrounded by larger regions of hypoperfusion including BA 22 and BA 37. Patients were tested daily on oral naming and spoken comprehension and scanned pre- and post-intervention to pharmacologically improve blood pressure. Time-to-peak (TTP) was measured and compared with TTP in the contralateral (right) hemisphere in these two regions. Improvement in naming in the first 2-3 days poststroke was correlated with improved blood flow (to less than 2.5s delay relative to normal hemisphere) in all six patients. Use of serial PWI to assess temporarily dysfunctional tissue in acute stroke also provides converging evidence of brain/behavior relationships prior to functional reorganization, in this case demonstrating that BA 22 is essential to oral naming and comprehension.
More recently, in a study of 50 patients, Kannan, Kleinman, et al. (2007) demonstrated that deficits in word comprehension were most strongly associated with abnormal diffusion weighted imaging (DWI) or perfusion weighted imaging (PWI) in posterior superior temporal (BA 22) and parietal cortex (BA40), where blood flow was at least 4s delayed compared to homologous regions.
In a DWI/PWI study of 170 acute patients within 48 hours poststroke, Hillis et al. (2006) provided evidence that semantic errors in oral naming and/or comprehension can arise from damage or dysfunction of different brain regions. Specifically, damage or dysfunction in BA 22 or 21 predicted errors in both naming and comprehension, while hypoperfusion in BA 37 (sparing BA 21) predicted semantic errors in naming alone.
Studies of motor recovery have also demonstrated restoration of function following reperfusion of hypoperfused penumbral areas. For example, Kleiser et al. (2005) demonstrated a “PWI-DWI mismatch” (an area of impaired perfusion at risk of infarction) in two acute stroke patients. After Internal Carotid Artery (ICA) recanalization, language (for patient with LH infarct) and motor task-specific (for patient with RH infarct) activation could be seen utilizing BOLD fMRI in peri-infarct regions that were previously acutely hypoperfused. This study provides evidence that at-risk tissue can survive, regain functionality, and support poststroke recovery.
The (ICA) Balloon Occlusion technique was utilized to test neurobehavioral effects associated with induced changes in cerebral hemodynamics in 44 patients1 (Marshall et al., 2001). Variability in patients’ performance was monitored during a sustained attention task in which patients were required to estimate 10-13s time intervals with a button press. Three groups emerged after carotid occlusion: one group demonstrated normal absolute cerebral blood flow (CBF) while maintaining consistent performance; a second group experienced a drop in CBF along with deterioration in task performance that spontaneously returned to normal; a third group demonstrated the largest effect in lowered CBF and was unable to recover task performance until the carotid occlusion was reversed. These results suggest that some compensatory mechanisms related to functional reorganization may be occurring in the hyperacute state in the face of persistent hypoperfusion.
Patients without infarct, but with unilateral cerebral hypoperfusion in the absence of stroke, may also demonstrate functional reorganization2 (Marshall et al., 2006). Fourteen such patients were studied with fMRI BOLD during a motor task. Activations in nine of the 14 with abnormal vasomotor reactivity significantly differed from those of control subjects, showing ipsilateral motor activation in homologous regions in the non-hypoperfused hemisphere. These findings are significant in that they suggest that hypoperfusion alone, in the absence of demonstrable brain injury on anatomical imaing may be sufficient to induce brain reorganization.
Some patients can have normal function in spite of chronic hypoperfusion. Mast et al. (1995)3 studied a patient with arteriovenous malformation (AVM) who was administered a superselective anesthetic injection (WADA) during language testing. This patient demonstrated very low blood flow (25 mm Hg). In spite of ischemia, tissue was evidently normal, because the patient temporarily exhibited symptoms of Wernicke’s aphasia after the injection. In fact, many chronic aphasia patients demonstrate hypoperfusion that doesn’t move into infarct. With restored blood flow some do recover function, but it depends on the level of ischemia, area of brain, and timing.
It is critical to consider the ongoing dynamic state of the vasculature in this equation. Use of multimodal imaging techniques can provide critical diagnostic information that may identify salvageable penumbral tissue in the early hours and days poststroke. These methods can also provide converging evidence regarding necessary and sufficient structure/function relationships in the brain.
Can fMRI, TMS, NIRS, etc complement the assessment of diffusion-perfusion MR imaging, DTI or spectroscopy for viable tissue that may be available to subserve later recovery?
It is possible that ischemia and the PWI-DWI mismatch may be part of a neuroprotective mechanism, but studying task-related activation in patients in the hyperacute phase is fraught with difficulties. Not many studies have attempted to examine patients in the early hours poststroke. Kidwell examined 18 patients within the first six hours, but in 16 of the patients massive head movement rendered the results uninterpretable.
Questions also remain concerning the implications of chronic low blood flow with regard to imaging methods. It is unknown, for example, whether the BOLD signal is normal in global hypoperfusion. It has been demonstrated to be undistinguishable from that of control subjects (Krakauer et al Ann Neurol 2004, Chmayssani et al Neurology 2007)4, 5.However, other studies (e.g., Roc et al., 2006)6 suggest that BOLD activation in patients with altered CBF may be quite different from healthy controls in spite of equivalent task performance on a simple visuomotor task. Roc and her colleagues demonstrated characteristically different hemodynamic response functions (HRF) in primary motor cortex in patients with hemodynamically significant stenoses, compared with neurologically healthy controls, in spite of no differences in the motor performance between groups.
The question remains whether or not blood flow is a good marker of neural activity in these patients. In rats, most show a BOLD response but with delayed peak, e.g., 6-8 s. In addition, in the contralateral hemisphere, the signal may be accelerated when one carotid is tied off. Since the BOLD response is associated with recruitment of additional blood flow, and blood flow may be compromised, it is advisable to examine individual patients’ HRF, rather than to rely on the canonical HRF during statistical analysis. Even normal subjects demonstrate different HRF between subjects, between regions, and over time. Patients’ HRF in particular can be expected to evolve over time.
Perhaps the most illustrative case is a patient who had recurrent transient ischemic attacks (TIAs) and no activation in ipsilateral motor cortex although he appeared behaviorally normal. It is possible that lack of activation in some patients during normal behavioral performance is related to impaired vasomotor reactivity, rather than carotid stenosis per se. This could create a lack of sensitivity to the BOLD response that is not related to hypoperfusion in the absolute sense, but rather to the fact that increased blood during the resting state changes the relative amplitude. Migraine studies have also shown a reduced BOLD response during flashing checkerboard paradigms.To assess the impact of hypoperfusion it may be better to evaluate the activity in the opposite, non-hypoperfused hemisphere, shown to be increased in the setting of chronic carotid disease (Marshall et al JCBFM 2006)2 and to return to normal in the setting of revascularization (Chmayssani et al Neurol2007(abs))4.
It is also possible that accuracy in localization may be “off” in terms of so-called perilesional activation, e.g., whether the region selected by BOLD response is related to maximum neural activation or whether it may be cms away from the actual centroid of neural activation. It may possibly be related to the degree of latency in TTP. Perilesional activation could also just reflect restored blood flow in the region, rather than a marker of neural activation. It’s an empirical question, but the study hasn’t been done.
Other methods have been used to predict recovery including examining secondary degeneration using diffusion tensor MRI ([DTI]; e.g., Liang et al., 2007)7. Twelve patients w/ subcortical infarct involving the posterior limb of the internal capsule underwent DTI at 1, 4, and 12 weeks post-stroke. Correlations between per cent changes in DTI measures (mean diffusivity and fractional anisotropy [FA]) and clinical scores (NIH Stroke Scale, Fugl-Meyer scale and Barthel index) were assessed. Low FA values were correlated with relatively impaired recovery, suggesting that FA may be useful as a physiological marker for recovery. Questions remain, however, regarding individual vs. group level predictability and the timecourse of degeneration, and the degree to which edema may change the amount of free water effecting FA. These questions hamper the current clinical usefulness of DTI.
Magnetoencephalography (MEG) is another non-hemodynamic dependent measure that has demonstrated the ability to predict longterm effective recovery (Tecchio et al., 2007)8. Thirty-two patients were studied at two timepoints: acutely and once stabilized. Clinical outcomes were grouped at three levels: worsening, partial recovery, and complete recovery. Two variables were observed to predict recovery or worsening of symptoms. In addition to lesion volume, an increase in delta activity in the contralateral hemisphere was predictive of worsening of symptoms. It was hypothesized that the contralateral correlation may be due to neuromodulation, i.e., the perilesional region’s release of inhibitory effects on the unaffected homologous region. EEG could also be used to obtain similar delta range data. Although this is speculative, there are plans to improve on this study by refining the electrophysiological data and providing more integration with the anatomical data.
These studies demonstrate potential clinical applications for predicting early poststroke patterns of activation (or physiological markers) that map onto later functional recovery or impairment. Currently, a lot of studies exclude patients with hypoperfusion due to imaging issues but these patients make up approximately 15% of stroke patients. Perhaps this exclusionary criteria needs to be reconsidered. At the same time, the BOLD response and its relationship to hypoperfusion is not well understood. Therefore, converging evidence from different imaging methods may help to assess the validity of the coupling between neural activity, blood flow, and the BOLD response. One could compare FDG PET to BOLD activation, or alternatively, one could use a modality that doesn’t depend on blood flow, e.g., DTI, MEG or EEG and then compare and contrast between modalities. Multiple collection of data over the timecourse of evolution of the stroke is recommended. Examination of individual patients’ HRF is also recommended, rather than relying on the canonical HRF during statistical analysis.
Alternative methods in fMRI analysis
There may be a performance confound when comparing patients to controls. When looking at the relationship between recovery and activation, the performance-related activation can’t be disambiguated from reorganization per se. Often statistical localization of activations is abused by subjecting it to reverse inference, i.e., “brain area, x, was activated, therefore cognitive process, y, was engaged”. Brain imaging can be useful even without statistical localization: 1) to test for existence of brain-behavior correlations; 2) to test for cognitive or other physiologic dissociations (via qualitative differences); or 3) to predict behavior, clinical diagnoses, or outcomes.
Multivariate testing in neuroimaging is not valid when the number of voxels is greater than the number of subjects. Using the multivariate linear model (MLM; Worsely et al., 1997)9, sensitivity for detection of diffuse effects is much greater than for localization. The typical trade-off between specificity and sensitivity is reflected in the power of localization vs. detection. The MLM uses a global test, i.e., testing all voxels simultaneously. It is much more sensitive to spatially diffused effects.
Analyses often rely on qualitative differences, e.g., we decide whether two conditions (or groups) engage the same or different brain circuits by examining whether the same (suprathreshold) blobs appear. But there is no formal statistical test for inferencing “blob pattern similarity”. What if the same circuit is engaged but simply to a greater degree in one condition than another? Typically, we would miss this pattern, because we would be missing all the information below the (arbitrary) spm threshold.
The MLM can test whether activation patterns in stroke patients are qualitatively different from controls, e.g., by testing for qualitative difference between left and right motor/sensory activation, or a squeeze test at different force rates. For example, one experiment asked if there were correlations between motor recovery at 3 months and brain activation in the first 48 hours poststroke (Krakauer, Zarahn, Lazar, Marshall 2007, unpublished data). To avoid the performance confound, motor recovery was represented as a change in score, i.e., the difference between the 3 month motor score minus the acute 48 hr motor score (Fugl-Meyer scale). Analysis included a 48 hr motor score covariate. The MLM tested whether the best combination of voxels to predict change does better than what is expected under the null hypothesis. Results detected a correlation between the delta score and activation. The pattern of recovery was qualitatively different from the motor execution pattern. The authors suggest that neural plasticity underlying recovery doesn’t simply involve an amplification of the motor execution network. In a traditional analysis, very few voxels would have survived the spm threshold for correction of multiple comparisons, in spite of a very strong global effect. Instead the MLM analysis yielded a prediction of the recovery pattern, accounting for 87% of the variance. Patients were included who were unable to move their affected hand at all. They weren’t outliers, but rather fit well into the curve. The unaffected hand also showed correlation with recovery. It is possible that derangements in CBF could have contributed to the conclusions of the study, but simply an overall lowered CBF would not have. Patients with large vessel disease were excluded. There are no claims being made about the mechanism underlying the recovery pattern, simply that the pattern is qualitatively different from that of normal subjects.
1. Marshall, R. S. et al. Recovery of brain function during induced cerebral hypoperfusion. Brain 124, 1208-17 (2001).
2. Marshall, R. S. et al. Hemodynamic impairment as a stimulus for functional brain reorganization. J Cereb Blood Flow Metab 26, 1256-62 (2006).
3. Mast, H. et al. 'Steal' is an unestablished mechanism for the clinical presentation of cerebral arteriovenous malformations. Stroke 26, 1215-20 (1995).
4. Chmayssani M, B. A., Handy C, Hirsch J, Marshall RS. Recovery of cerebral hemodynamics induces normalization of atypical ipsilateral motor activity on fMRI. Neurology 68(suppl), A65 (2007).
5. Krakauer, J. W. et al. Hypoperfusion without stroke alters motor activation in the opposite hemisphere. Ann Neurol 56, 796-802 (2004).
6. Roc, A. C. et al. Altered hemodynamics and regional cerebral blood flow in patients with hemodynamically significant stenoses. Stroke 37, 382-7 (2006).
7. Liang, Z. et al. A prospective study of secondary degeneration following subcortical infarction using diffusion tensor imaging. J Neurol Neurosurg Psychiatry 78, 581-6 (2007).
8. Tecchio, F. et al. Rhythmic brain activity at rest from rolandic areas in acute mono-hemispheric stroke: a magnetoencephalographic study. Neuroimage 28, 72-83 (2005).
9. Worsley, K. J., Poline, J. B., Friston, K. J. & Evans, A. C. Characterizing the response of PET and fMRI data using multivariate linear models. Neuroimage 6, 305-19 (1997).
5. Session Title: Diffusion Tensor Imaging (DTI)
Facilitators: Jennifer Newton, Rüdiger Seitz
Summarizer: Carlos Marquez de la Plata
The aim of the discussion was to determine how and to what extent DTI can contribute to research on neurologic impairment, recovery, and rehabilitation. Possible contributions of DTI are the prediction of recovery (prognosis), and the identification of theraputic approaches to maximize recovery given the damage to connections observed with DTI.
Overview of DTI
DTI allows one to examine the diffusivity of water to inform us about the directionality and integrity of white matter, as water diffuses anisotropically (greater diffusion occurs along particular directions) along intact axons, and diffuses relatively isotropically (diffusion occurs equally in all directions) in grey matter or injured white matter. Scalar measures about the diffusion characteristics in each voxel of the brain can be calculated from the diffusion tensor. Mean diffusivity (MD) is a directionally averaged measure of diffusion; whereas, fractional anisotropy (FA) is a measure of how much the diffusion tensor deviates from the isotropic form. One can compare these measures for specified regions of interest or using a voxel-wise approach.
Tractography, a technique that uses information about the predominant direction of diffusion in each voxel to reconstruct fiber trajectories in white matter, was discussed as a useful, albeit relatively user-dependent means of investigating white matter integrity. Though the tracts reconstructed by this method are often referred to as fibers/tubules. it is important to note that streamlines or pathways derived with tractography do not represent individual fibers. This is a consequence of the large size of imaging voxels (usually 1-3 mm in each dimension) relative to individual axons; a single imaging voxel could contain several thousand axons. There are various candidate parameters of structural integrity that can be extracted following reconstruction of tracts. These include total streamline count between target regions, as well as streamline density (i.e., streamline counts per voxel) and mean FA across the reconstructed tract volume.
Methods of Tract Reconstruction
Three dimensional tract reconstruction can be performed using linear (deterministic) or distributed (probabilistic) methods. Linear methods, including Fiber Assignment by Continuous Tracking (FACT), which attempts to determine connectivity utilizing the principal eigenvector of the diffusion tensor to provide a propagation direction for each voxel along a path. The limitation of this approach is that it produces only one reconstructed projection per seed point, when there could be multiple projections that are not accounted for. In addition, these methods do not take into account the uncertainty of the direction information in each voxel, which may lead to generation of erroneous trajectories. This uncertainty information, in the form of probability density functions (PDFs), is incorporated into probabilistic tractography methods, such as the Probabilistic Index of Connectivity (PICo) or the fMRIB Diffusion toolbox. The advantage of these methods is that they can provide a measure of confidence for the connections established, whereas linear tracking methods do not. However, a limitation to these methods is the difficulty in adequately modeling diffusion, and its uncertainty, in each voxel. Modeling areas known to contain multiple tracts propagating in multiple directions is a challenge for three dimensional tract reconstruction methods. However, there are techniques available that model a mixture of diffusion directions to represent intravoxel heterogeneity seen in areas such as crossing fibers and branches that sprout from white matter bundles reaching target cortex.
How Well Validated is DTI as a Measure of Structural Connectivity
DTI has shown concurrent validity as a measure of structural connectivity in stroke. For instance, Judith Schaechter presented tractography work by the Martinos Center showing the number of fibers seen in the ipsilesional cortical-spinal tract (CST) of patients with hemiparetic stroke is highly correlated with ipsilesional peduncular area; such that a lower fiber count in the CST corresponds to lower peduncular area as seen in morphemetric analysis.
Recent findings from published literature were also discussed and consequently highlighted potential confounds in the interpretation of tractography as a measure of structural connectivity. The dependence of propagation of streamlines on anisotropic diffusion was raised as a major issue for the use of tractography in the presence of white matter damage, which can lead to reduced anisotropy. The influence of retrograde and anterograde changes on the ability to reconstruct tracts was also discussed.
More general issues concerning the importance of seed point selection and incorporation of target areas into tractography algorithms were raised. J. Schaechter pointed out that tractography may result in false or aberrant projections which results from the uncertainty in the data given by the partial volume effect. Thus, restriction of the solutions by seed and target points is mandatory and an exploratory analysis driven by the DTI data is hardly possible. Dorothee Saur presented group analyses of intrahemispheric connections in the language system; the use of voxelwise t-tests of tractography-derived probabilistic maps of connectivity at the group level may reduce the influence of false or aberrant streamlines in the reconstruction of white matter tracts across subjects.
As with any technology, there are challenges to tractography. For instance, the reconstruction of white matter tracts is user dependent which may introduce error or bias and is dependent on the user’s knowledge of neuroanatomy. Insufficient knowledge of neuroanatomy and knowledge of structural connectivity can result in erroneous tracking. Another challenge is determining when it is appropriate to terminate a streamline in the brain. In many tractography algorithms this is determined by FA; however, this may be problematic in injured brains, as FA is reduced in white matter lesions and the reconstruction of tracts terminate far short of where the tract terminates in uninjured brains.
DTI was discussed as a measure with potential utility in combination with fMRI. Such a study might attempt to determine whether the start or end points of damaged white matter tracts correlate spatially with altered BOLD responses. The combination of structural and functional imaging modalities may be a useful method of assessing whether the degree of functional plasticity is related to the degree of structural connectivity.
6. Session title: Methods for Assessing Functional Connectivity: TMS, ERP, MEG, Statistical Analyses
Name of facilitators: Small and Cohen (Cohen Absent)
Summarizer: Monica Perez and Andy Mayer
Proposed Discussion Questions:
What are the pros and cons of these methods for assessing functional connectivity?
How sensitive are these methods to learning-based changes in connectivity?
Functional connectivity must be distinguished from anatomical connectivity.
Diffusion tensor imaging (DTI) is a technique based on diffusion weighted imaging (DWI) that allows an evaluation of the integrity of white matter tracts by virtue of its ability to visualize water diffusion along axonal pathways. Fiber tractography (FT), which is a three-dimensional visualized derivative of DTI data, permits visualization of the architecture and integrity of neuronal tracts.
DWI is more sensitive to underlying microstructural events of water molecules in biologic tissues than conventional MRI, which is largely limited to the macroscopic assessment of cortical and subcortical structures.
By examining directional diffusion, DTI allows the evaluation of the orientation of white matter fibers determined from the primary vector of the diffusion tensor and can show the three-dimensional FT.
Quantitative measures such as mean diffusivity and fractional anisotropy are available with DTI. These measurements are not available from conventional MRI.
DTI can be used to visualize the white matter pathway, even prior to myelination. On the other hand, conventional T1- and T2-weighted signal intensity changes in white matter are strongly dependent on the presence of myelin.
Allows the probing of direct physiological connections when combined with other techniques such as TMS (M1 is stimulated and the corresponding motor response recorded in contralateral hand).
FT. Poor specificity in terms of identifying the type of fibers that are quantified.
FT is also not reliable in regions where there are several fiber crossings.
DTI provides information concerning the average orientation of fibers at the voxel level, and if this volume-averaged information is used to reconstruct a pathway, false positive projections may be observed.
By comparison with the structural concept of anatomical connectivity is the more functionally relevant concept of functional connectivity.
Functional connectivity refers to correlative relationships that might exist between the activations of distinct and often well separated neuronal populations, without any reference to physical connections or an underlying causal model. In contrast, analyses of effective connectivity are based on statistical models that make anatomically based assumptions and limit their inferences to networks comprising a number of pre-selected regions. Sometimes these two terms are used interchangeably. The electrophysiological techniques (MEG/EEG) have superior temporal resolution and are able to measure functional or effective connectivity in ways that have greater physiological meaning (rather than just statistical meaning), but suffer from the inverse problem (i.e., the fact that any single electrophysiological pattern can be generated in an infinite number of ways, thus making it difficult to infer the sources without additional constraints or information). Functional connectivity can be defined in terms of activity that occurs during a given cognitive/motor task or during “rest” (the absence of a specific task) and is measured with several different analysis techniques including seed-voxel analyses, independent component analyses (ICA), dynamic causal models, structural equation modeling, granger causality, etc.
Dynamic causal models (DCMs) incorporate a model of the neuronal level that aims to relate fMRI activity to theoretical activity at the neuronal level.
DCM uses Bayesian model comparison; therefore, one can compare non-nested network models.
In contrast to the majority of univariate statistics, functional connectivity analyses are not dependent on amplitude differences.
If functional connectivity is measured during rest, then one can potentially eliminate confounds between different patient groups that are a result of differences in behavioral performance.
There appears to be converging evidence from three different areas of imaging regarding a similar “default-mode” network (Raichle’s hypermetabolism at rest studies, Binder’s deactivation during a task (fMRI) and independent component analyses).
These fMRI methods are analyses of functional connectivity, which make no inferences about the actual anatomy of the connections involved.
A current limitation of DCM is that model fitting is computationally demanding.
Another limitation of DCM is that neuron-dynamics in each region are characterized by a single state variable (neuronal activity). Thus, the method does not incorporate a model of different neurotransmitter systems.
Normal-appearing functional connectivity does not necessarily mean that normal coupling between brain regions exists.
One prominent confound in functional connectivity measurements is physiological noise in BOLD signal activity, which can influence correlation measurements.
It is not clear what “rest” really means in terms of underlying cognitive processes and whether this is just another task.
The relationship between neurovascular uncoupling following a stroke and how this affects the BOLD response has not been determined.
During motor learning processes, there are changes in neuronal network activity, which may be partially related to the development of new synaptic connections or to the unmasking of silent synapses. Although imaging methods can be sensitive enough to detect changes in neuronal network activity before and after an intervention, the nature of these changes requires careful interpretation. Functional connectivity is a measurement of the spatiotemporal synchrony or correlations of the BOLD fMRI signal between anatomically distinct brain regions of cerebral cortex. In the resting state, low-frequency fluctuations of the BOLD signal, which are related to neuronal spontaneous activity, can be used to identify functional connectivity among different brain regions. Then, the study of functional connectivity using resting state fMRI data may provide an opportunity to detect different aspects of plastic changes associated to motor learning processes, such as the appearance of compensatory plasticity; increased functional connectivity could indicate the dominance of a compensatory mechanism. However, one of the problems of using functional connectivity studies based on resting state fMRI is how to interpret these results in terms of active task performance.
Summary of the discussion:
Connectivity research can generally be classified into imaging techniques that are based on anatomical connectivity (DTI, fiber tracking) and those that are based on functional (fMRI, EEG and MEG) or effective (EEG and MEG) connectivity. The primary benefit of anatomical connectivity studies is that the underlying anatomy and physiology of the system is well characterized. Functional connectivity may hold great promise for understanding of network disruptions in the brain following trauma to the central nervous system, but more works needs to be done to quantify what these different techniques are measuring. Perhaps the single greatest benefit of true resting state data is that the clinician does not have to be concerned about differences in behavioral performance that necessarily confound studies examining functional activation during a task. Perhaps the greatest weakness of current connectivity techniques is the uncertainty regarding underlying physiological mechanism (i.e., what is rest?), the reliability of connectivity analyses and subsequent results, and the potential differences in connectivity in local penumbral/lesioned tissues compared to more unaffected regions following trauma. Finally, both fMRI and MEG/EEG techniques have their respective limitations in either being able to measure effective connectivity (MEG but not fMRI) or having ill-posed solutions for localization (MEG but not fMRI).
Synthesis / Recommendations:
Overall in this session, different imaging methods to study anatomical and functional connectivity between brain regions were described. Network models are important tools that can provide a common framework for describing connectivity of distinct brain areas at the level of anatomy and function. However, currently, there is no consensus on the most accurate or efficient method of detecting or measuring functional connectivity using fMRI. Importantly, during interpretation of the results, it is critical to consider that in all cases, even for anatomy, the network descriptions are only approximations of the real systems.
7. Session Title: Imaging Pharmacologic Modulation of Neural Activity in Diffuse Brain Injury: BOLD and Perfusion MRI, ERP, MEG
Facilitators: Francois Chollet and John Whyte
Summarizer: Anthony Chen
Proposed Discussion Questions:
What are the pros and cons of these methods in distinguishing between cognitive/motor effects of drugs vs. direct vasoactive effects?
How do drug-induced changes in activation relate to drug-induced performance changes?
Can baseline imaging results serve as predictors of drug response?
Can imaging markers of drug response serve as surrogate outcomes in screening for clinically useful drugs?
Summary of the discussion
Overall question raised by this symposium: What are some of the issues that need to be considered when designing studies that examine pharmacologic modulation of neural activity?
Introduction of general issues by Dr. John Whyte: Studies that strive to examine pharmacologic effects with imaging face a range of challenges that differ from strictly task-based functional imaging studies. Some of the questions to consider include:
Repeatability of measurements over time.
Distinguishing the primary effects of drug on vasculature from neural effects. The pharmacologic systems of greatest interest for rehabilitation studies are those that are involved in regulating vasculature—acetylcholine, dopamine and norepinephrine are among these.
What is the study question? Two important general objectives are to better understand mechanisms of treatment and to predict the effects of drug on performance and/or recovery.
What are the appropriate study designs for pharmacologic imaging studies?
- The nature of the scientific question to be pursued needs to be defined, leading to different study designs. Some studies may have the objective of using imaging data to predict recovery or the likelihood of long-term benefits of a drug, while other studies may have the objective of understanding the mechanisms by which a drug acts to alter rehabilitation-relevant neural processes.
- The distinction between clinical trial design and designs appropriate for mechanistic studies should be considered. At the current time, imaging methods may be considered most appropriate for providing data for mechanistic questions, and are not necessarily appropriate for addressing clinical trial-type questions. Standard clinical trial designs may not be the most appropriate starting point for most imaging studies.
- As one important goal of pharmacotherapy in rehabilitation is to augment the effects of other interventions, study designs should reflect this. Bruce Dobkin discussed this as an ‘enrichment’ strategy. An important scientific question is how a drug changes rehabilitation-induced learning processes. In these scenarios, it may be important to design studies that examine neural changes during the course of therapy, as just measuring the baseline compared to post-treatment outcome may provide information on the consequences of treatment, but not necessarily the effects of treatment.
- Dr. Chollet) presented data from a series of studies that utilized double-blind placebo-controlled designs, with fMRI and TMS measurements. Significant questions were raised by this discussion, regarding the best study designs, optimal dosing, and appropriate outcome measures.
What are some dosage considerations when designing pharmacologic imaging studies?
- Basic studies utilizing a range of doses in order to determine dose response effects may be needed.
- In mechanistic studies, an important objective is to determine neural changes that correspond to successful treatments (that is, treatments that alter clinical variables). The dose at which this occurs is likely to be different for different individuals; therefore, it would be logical to include individualized dosing as part of the study design.
- Dr. Chollet presented a series of studies highlighting the following key points of general interest:
All significant clinical trials have until now been negative, including those of amphetamine compounds.
The published series have shown that drug-induced performance increases can be correlated with changes in the activity of specific cerebral networks. This is now demonstrated with both fMRI (BOLD changes) and evoked potential technique (TMS) and it suggests that neuroimaging tools might be used as surrogate markers associated with clinical endpoints in clinical trials including a limited number of patients.
The effects of chronic drug administration may be very different from those of acute single doses. We need to develop trials with chronic doses.
Clinical trials associating drug-induced clinical performance and fMRI and/or TMS changes in recovering patients to date have tested acute performance but never learning or re-learning.
What are appropriate (neurophysiologic) measurements for pharmacologic imaging studies?
- Imaging methodology is a particularly important consideration for drug studies. Significant discussion of perfusion methods in preference to BOLD MRI were raised by John Detre as well as John Whyte, since the availability of a scalar measure of blood flow allows repeated testing in different drug conditions, and allows one to distinguish between primary vasoactive effects (by measuring baseline flow) and cognitively induced changes in flow.
- The issue of variability—is it possible that an effect of a drug is to alter variability (in physiology or neural responses) itself?
- Should anatomical outcome variables be considered, given the inherent variability in neurophysiologic variables?
- What are the most appropriate behavioral endpoints for a particular study? The dynamic range, repeatability and appropriateness of the endpoints for the hypothesized mechanisms of action should be considered. Furthermore, more fundamental studies that examine these endpoints may need to be done before embarking on a pharmacology imaging study.
- Dr. Junghoon Kim presented data from a study using CASL perfusion imaging to evaluate effects of methylphenidate. Key points of general interest included: determination of drug effects on a ‘baseline’ or ‘resting’ state in order to better interpret changes on task-related activity; the use of quantitative perfusion imaging to improve reliability of serial measurements, as above.
Can baseline imaging results serve as predictors of drug response?
- Dr. Steve Cramer presented data addressing the question of whether baseline measurements may predict the effects of intervention at a later time point, highlighting the following key points of general interest: the amount of improvement in brain function is generally likely to be related to baseline functioning, for example, lower motor cortex signal may suggest a higher capacity for improvement.
How do drug-induced changes in activation relate to drug-induced performance changes?
- It was generally agreed that examining the relationship between drug-induced neural changes and behavioral changes is particularly important in interpreting the significance of the measured neural changes. However, it was also argued that neural changes that occur in the absence of behavioral changes may also be particularly revealing regarding mechanisms of drug effects, since they lack the “performance confound” (the fact that distinctly different qualities and strategies of performance are always presumed to be associated with distinct patterns of neural activity) that plagues functional imaging research. For example, if changes in baseline regional CBF with drug predict improvements in active performance, the imaging changes cannot be confounded with that performance.
Can imaging be used for surrogate markers in drug studies?
- This was discussed only briefly, and no data were specifically presented or discussed that support this potential use of imaging as surrogate markers for clinical outcomes in drug studies. Indeed, extensive studies with this particular issue in mind would need to be done in order to establish specific imaging markers as surrogate markers. Furthermore, many of the participants would likely agree that the most important outcomes are clinical/behavioral outcomes. As repeated in a number of discussions, imaging may be most valuable for determining mechanisms of treatment effects, and potentially providing markers that may be predictive of treatment effects.
Although specific consensus was not reached, the following were significant questions or themes that arose from the discussion:
- Study designs for pharmacologic imaging studies may require tailoring in order to best address the primary questions, which includes questions of mechanism or prediction, but rarely proof of efficacy. Thus, double-blind randomized placebo controlled designs utilizing a uniform standard dosing, although the ‘gold standard’ for clinical trials, may not be the best design depending on the study objectives.
- Qualities of the measurement methods need to be considered carefully, including inherent variability in the measurements for serial study designs and sensitivity of the measurements for processes most likely to be modulated by the chosen drugs.
1-A single dose of the serotonin neurotransmission agonist paroxetine
enhances motor output: double-blind, placebo-controlled, fMRI study in
healthy subjects. Loubinoux I, Pariente J, Boulanouar K, Carel C, Manelfe
C, Rascol O, Celsis P, Chollet F. NeuroImage, 2002, 15, 26-36.
2- Fluoxetine modulates motor performance and cerebral activation of
patients recovering from stroke. Pariente J, Loubinoux I, Carel C, Albucher
JF, Leger A, Manelfe C, Rascol O, Chollet F. Ann.Neurol 2001, 50, 718-729.
8. General Discussion
Facilitators: Nick Ward and Myrna Schwartz
Summarizer: Steven Jax and Sarah Shomstein
The main framework of the closing discussion section was to consider the rationale and methods for conducting neuroimaging studies in the service of understanding mechanisms of recovery and of therapeutic interventions.
Rationale for rehab-inspired neuroimaging research
The consensus appeared to be that most early studies using functional imaging techniques were primarily observational. However, we are now entering the new and exciting phase of neuro-rehabilitative research in which is hypothesis driven.
The discussion progressed to ask whether hypothesis driven research is beneficial to the rehabilitative field. The point was made that a pragmatic approach to treatment will always be required. However, a distinction needs to be made between treatments which aim to help adaptation to impairment and those which aim to minimize the impairment itself. The former are well established and form the cornerstone of neurorehabilitation. The latter are less well developed, but it is in this field that neuroscience can help further understanding. In this framework, basic research is imperative for understanding the mechanisms of cognitive control and can later be used to develop rehabilitative techniques designed to minimize impairment. However, the current practice of neurorehabilitation will continue to be pragmatic until evidence suggests that changes in practice should be made.
The consensus appeared to be that both approaches are worthwhile. There is great value in both – having a clear theoretical notion of your treatment as well as having a treatment that works and adding a neuroimaging component to identify possible changes at the neural level. Whichever approach is taken, it is critical to have a research question that matches both the hypothesis and the method used to investigate that question.
Understanding the mechanisms
An important issue for future work will be to better understand the similarities or differences in neural changes following spontaneous recovery and following rehabilitation. Thus far, it is unclear whether rehabilitation techniques affect the brain in the same way as spontaneous recovery or if it is possible for rehabilitation to change the way recovery progresses. Future work should consider this issue.
Prediction (e.g., to predict the best way to deliver treatment)
One of the key themes has been that patients with brain injury have as many differences as similarities from one another. It is unlikely that a particular treatment will be equally effective in all patients as a result of differences in anatomical damage, time since damage, biological age, concurrent medication, genotype etc. Thus different patients might show different patterns of neural reorganization, either spontaneously or in response to treatments. If the goal of a treatment is to promote cerebral reorganization then structural and functional imaging might be able to predict whether an individual will respond to that particular treatment, thus allowing stratification of patients based on their likelihood of responding. However, at present the field is not yet mature enough to provide treatment prescriptions, but this is clearly a goal for future work.
Reliability of the tools and methods of analysis
The quality of the data acquired is critically important. Much care should be devoted to ensuring that the data is of the best quality possible. FMRI data sets are complex and it is worth remembering that they be analyzed in a number of ways. The following points were felt worthy of consideration.
First, and foremost, experimental design should be given the most important consideration when designing any kind of a neuroimaging study. The suggestion is to attempt to focus on simple paradigms in which most of the experimental factors can be rather well controlled. This will reduce the noise and increase explanatory power for observed effects.
Data may be contaminated by motion artifacts. It is suggested that most of the motion algorithms can filter out motion rather well. As long as movement is not task correlated (i.e., patient moves with every presentation of the stimulus) this motion can be filtered out effectively. However, there are methods (e.g. uwarp in SPM5) that are designed to deal with task correlated head movement. It was stressed that time spent making the subject comfortable in the scanning environment with adequate time to practice the task often helps reduce motion artifacts.
More sophisticated statistical analyses may be employed. We mostly use mass univariate statistical techniques to provide information about relative increases or decreases in activation in brain regions. This kind of analysis is certainly valuable, but may not pick up other kinds of changes such as those designed to make inferences about changes in connection strength between regions
Some considered that thresholding of data seems arbitrary. It was pointed out that the first level ‘contrast images’ from a single subjects that are used in a true mixed effects analyses (at group level) are not thresholded. When thresholding is used, it was pointed out that widely accepted methods and levels of thresholding are in use. This includes the use of correction for multiple comparisons and the option of restricting a search volume based on a priori anatomical hypotheses. It was suggested that an informative statistical measure is performing power analysis. This measure would be extremely useful in interpreting the results at any given threshold (i.e., power to threshold relationship).
Generalizability of the principles
It was agreed that different areas of neuro-rehabilitation seem to be at different stages of development, therefore the level of analysis that seems to be appropriate for one area might not be appropriate (or even feasible) for another. For example, research inspired by psycholinguistic theory has been very useful for understanding language related impairments after stroke, but a good understanding of TBI has been lacking. Also, stroke studies can go beyond descriptive methods because the physiological markers of stroke are better understood. In terms of the overall general method of approaching any single study, it was noted that the same standards cannot be applied to say stroke studies as to studies of TBI.
Barriers to collaboration
Several barriers to collaboration were identified:
Differences in equipment (i.e., different MR machines) and analysis software need to be considered. Whether data from different scanners can be combined is unclear, but studies assessing this possibility are underway. Thus this question needs to be positively addressed empirically.
Although most of the participants agreed that larger sample studies were needed, differences in research interest (e.g. language vs. motor abilities) and population of interest (stroke vs. TBI) make consensus about even the research question for multisite collaborations difficult. Collaborations therefore should be based on agreement about a particular research question and hypothesis to be tested.
In order to convince a large organization to fund a significant multicentre study, it would be very useful to have some pilot data to demonstrate that such a project is at least feasible.
Finally, it was again reiterated that some pilot funds for junior faculty are available from the NCRRN, and that proposals will be due in the few months following the conference.
© copyright 2006, LQJ