Meeting Summary

Optimizing fMRI Approaches to Adolescent Mental Disorders

August 17, 2006 – August 18, 2006
Rockville, Maryland

Sponsored by:
Division of Pediatric Translational Research and Treatment Development (DPTR)
Division of Neuroscience and Basic Behavioral Science (DNBBS)

On August 17–18, 2006, the National Institute of Mental Health sponsored a workshop that brought together researchers involved in the use of functional magnetic resonance imaging (fMRI) to study adolescent mental disorders and normal development, as well as scientists involved in integrating fMRI data with data from other imaging modalities. The goal was to address issues involved in such research toward the goal of optimizing study designs and approaches to improve our understanding of the neural bases of these disorders.

Major Themes

Listed below are some major themes and discussion points addressed during the workshop.

• Conceptual Challenges
• Adolescence as a Developmental Epoch
• fMRI Paradigms and Approaches
• Dependent fMRI measures
• Stability/reliability of fMRI activations
• Longitudinal designs
• Individual differences in fMRI activation patterns
• Performance issues
• Technological challenges for multi-site studies
• Arterial spin labeling (ASL)
• Diffusion tensor imaging (DTI)
• Magnetic resonance spectroscopy (MRS)
• Some useful approaches for future research
• Technology, tools, and resources — needs

Conceptual Challenges

• A need exists to distinguish epiphenomena and compensatory responses from core pathological processes.
• Cohort effects may confound age-related comparisons in cross-sectional studies of patients. Adults with persistent symptoms may differ in important ways from children with the same disorder.
• The correlational nature of imaging data necessitates its integration with other data and approaches in order to constrain interpretations and move closer to causality.

Adolescence as a Developmental Epoch

• Adolescence is characterized by a transition from exogenously-driven to endogenously-driven behavior (cognitive control).
• Structural imaging data indicate that the association areas in each lobe are the latest to mature; the underlying mechanisms this late maturation are thought to be synaptic pruning and myelination, which enable faster neurotransmission.
• fMRI data suggest greater prefrontal activation in adolescence than in either childhood or adulthood (i.e., a curvilinear relationship) and more widespread distribution of activation in adults relative to adolescents when using eye movement tasks to assess cognitive control mechanisms. The transition from adolescence to adulthood is marked by increased collaboration among brain regions.
• Researchers do not yet have a solid understanding of the effects of puberty and better metrics are needed. For example, the length of puberty (as a possible indicator of plasticity) and the domain of interest (e.g., motivation) may prove important.

fMRI Paradigms and Approaches

• Cognitive control and emotion regulation are domains of high clinical relevance to adolescent mental disorders, but neither are unitary processes. Cognitive control processes include manipulation of information (e.g., in working memory), task management (e.g., switching or multi-tasking), and response inhibition or selection. These processes engage different sub-regions of prefrontal cortex, involve different functional connectivity, and develop at different rates
• Interactions of age with condition and patterns of activation are better indicators of developmental change than are main effects involving age-related differences in levels of activation.
• Negative attentional biases may contribute to depression, but the inability to regulate or repair sad moods is more germane to the sustained negative affect that is a cardinal symptom of depression.
• Findings from fMRI paradigms with control subjects that involved sad mood induction followed by a mood repair manipulation — the recall of positive memories — suggested the involvement of the parahippocampal gyrus in the elaboration of sad mood and the subgenual anterior cingulate gyrus in the repair of sad mood.
• fMRI studies of adolescents at risk for depression by virtue of their family histories suggest aberrant amygdalar activations not seen in controls. Relationships to stress reactivity and genotype may be important.
• Interpretive value will be enhanced when information from fMRI is integrated with other biological and behavioral data.

Dependent fMRI measures

• As an alternative to level and/or extent of activation, patterns of activation, such as those captured in laterality indices for language, may be more developmentally sensitive, robust, and meaningful at an individual level. Relationships between prefrontal activations and other regions may be particularly relevant to adolescent mental disorders.
• Caution is, however, warranted when combining measures as one may also be increasing error variance.
• Variability in the localization of activations might be captured with measures of distance from the maxima of group-averaged activations. The focal extent and variability in localization might serve as a metric for comparing patients versus controls.
• There is a need to increase reliability of individual results in order for fMRI to be useful clinically.
• Whole brain multivariate techniques may be particularly useful for studying changes in patterns of brain activity.

Stability/reliability of fMRI activations

• Stable, reliable fMRI markers may prove useful in tracking emotional reactivity, as well as therapeutic responses (e.g., to medications).
• Paradigms employing fearful facial expressions have proven robust in activating the amygdala in adults. However, in children, these activations are less reliably obtained when using neutral faces as a baseline condition and may reflect arousal more than valence.
• State anxiety may track the response to neutral (i.e., potentially ambiguous) faces in adults and children.
• Arousal and valence may be separable, with dorsal amygdala responses reflecting arousal relatively more than the ventral amygdala which tracks valence.

Longitudinal designs

• There are few fMRI studies of test-retest reliability, and it is unknown whether fMRI data from adolescents are less stable than adults over time.
• Sources of variance include age-related anatomical differences, physiological variables (arousal, emotional state), time of day, site differences including personnel, and other contextual influences, all of which could be standardized in a protocol.
• Analysis parameters (percent signal change versus Z-score; region of interest (ROI) versus voxel-wise approaches; degree of smoothing) will influence stability and affect sensitivity. Measures that are highly sensitive to threshold levels, such as cluster extent, are unlikely to be stable or reliable.
• Some research has shown moderate reliability over 8 weeks when using percent signal change in an ROI as a measure. Z-scores were less reliable. Reliability was improved by defining an ROI based on a region of activation at time 1, as compared to an ROI defined by a Talairach atlas.
• The process of teasing apart signal components in independent components analysis (ICA) may function to reduce noise and increase reliability.
• Combining fMRI with other imaging measures (diffusion tensor imaging, perfusion) and physiological covariates (skin conductance response, pupil dilation, respiration, etc.) may reduce variability and increase reliability.

Individual differences in fMRI activation patterns

• Understanding individual variation is important for studies of mental health.
• Approaches include correlating activation patterns with abilities (e.g., neuropsychological) or traits (e.g., personality) measured outside of the scanner, using multiple regression to control for other pertinent factors.
• Network analyses that are hypothesis- or data-driven, such as structural equation modeling, may be particularly useful.
• Developing sensitive and valid tasks for use in psychiatry is critical to any potential clinical applications. Researchers need to determine how frequently a task results in a specific activation in normal subjects in order to define abnormalities in a disorder.
• Performance measures are important for verifying compliance.
• Temporal stability of the imaging and performance data provides a platform for potential sensitivity to diagnostic differences and treatment effects.
• Activation data from an fMRI study employing an attentional task designed to activate the dorsal anterior cingulate was presented as an example. Several studies of ADHD have reported abnormalities in this region which have responded to stimulant treatment.

Performance issues

• Activations can be examined to determine which are specifically performance- versus age-related.
• When comparing patients to controls or different age groups, differences in processing strategies, arousal, effort, frustration, and emotional reactions to one's performance may interact with diagnosis or age, even in the absence of objectively-measured performance differences. Elementary, simple tasks may minimize these effects.
• Designing tasks that tap the same construct across different age groups remains a challenge for developmental studies.

Technological challenges for multi-site studies

• The Function BIRN (Biomedical Informatics Research Network) is a consortium of researchers who are taking on the challenge of combining fMRI datasets from different sites and creating capabilities for large multi-site neuroimaging studies. Site differences (e.g., scanners, procedures) have been shown to exceed inter-subject differences.
• Methods used to reduce site differences include use of an agar phantom, a breath hold scan as a surrogate for a maximum BOLD response, standardization of response devices, tasks, and approaches to data analysis, and various calibration strategies. While progress has been made, much remains to be done.

Arterial spin labeling (ASL)

• ASL uses an endogenous flow tracer analogous to O-15 water and can yield a quantitative measure of cerebral blood flow.
• Developments over the past decade now permit efficient, noninvasive whole brain imaging, appropriate for use in studies of development.
• Resting perfusion fMRI studies may offer a useful alternative or adjunct to BOLD fMRI studies. Resting perfusion may be correlated with diagnosis or behavioral measures obtained outside the scanner.
• Activation signals in ASL come primarily from the microvasculature (capillaries), as in PET.
• ASL may offer greater stability for developmental studies and studies of phenomena that cannot be easily turned on and off, (e.g., mood states, learning, and pharmacological effects).
• ASL may have potential as a biomarker for abnormal brain function due to its stability.

Diffusion tensor imaging (DTI)

• DTI is a noninvasive technique for mapping information on the coherence and direction of white matter fibers in the brain.
• There is increasing interest in combining fMRI and DTI to map connections between regions of activation by using the functional activation as a seeding point for DTI fiber reconstructions.
• Conceptually, this raises several issues: (1) fMRI activations are limited to gray matter; DTI is limited to white matter; (2) seeding points are highly sensitive to arbitrary fMRI thresholding; and (3) the angle of intersection between fMRI and DTI eigenvectors can vary such that caution is needed to avoid propagating connections in erroneous directions.
• Potential solutions include developing probabilistic DTI models and over-sampling.
• There remains a fundamental lack of understanding of the relationship between what we see at the MRI level, at the diffusional MRI level contrast, and the microscopic structure of the tissue.
• Comparison of the amount of a marker for intracellular space, e.g., N-acetyl aspartate (NAA), may help distinguish intra- and extra-cellular DTI signals.
• Decreased fractional anisotropy, as an indication of the structural integrity of fibers, may reflect fewer axons or less myelin; a finding of less NAA might help constrain the interpretation of the data.
• Similar ambiguity characterizes other imaging measures, such as cortical thickness.
• To overcome the correlational nature of MR, longitudinal causational validation studies involving animals are needed.

Magnetic resonance spectroscopy (MRS)

• MRS measures neurotransmitter metabolism and can be used to examine the neural basis of the fMRI signal by correlating changes in neuroenergetics — cerebral metabolic rate of oxygen (CMRO2) — with neural activity and estimating the fraction of total brain activity accounted for by changes in a task.
• A large proportion of the energy costs of neuronal activity are related to glutamate release. Modern models of neuroenergetics support the view that functional activity accounts for 85 percent of brain energy metabolism. The relationship between glutamate release and CMRO2 is linear and provides an index of neuronal signaling.
• Changes in fMRI signal are proportional to changes in neuroenergetics (and thus, neural activity).
• The amount of brain activity in the resting state is many times greater than the change in signal induced by a particular task.
• Researchers have used MRS to examine changes associated with disease:
  • Disturbances in glutamate flux between neurons and glia can lead to excitotoxicity and impaired neurotransmission.
  • Lower concentrations of GABA have been detected in depression, while certain treatments result in increases in GABA.
• MRS methodology has limitations. For example, it is unknown how glutamate metabolism changes during development.

Some useful approaches for future research

• There is a need for studies to integrate information across imaging modalities for improved understanding of biology at the level of the individual.
• Multimodal imaging and studies that combine imaging with randomized clinical trials can help overcome biased sampling, constrain interpretations of pathophysiology derived from fMRI, and elucidate mechanisms of treatment response. Preceding this, small studies of the best cohort possible are needed to see if one can detect therapeutic changes on imaging.
• Studies are needed to identify biomarkers and to determine which modalities will be most helpful for specific diseases or purposes.
• Coupling fMRI with perfusion and MRS to look at medication effects would provide better quantification, stronger links to chemical changes, and more generalizable information than BOLD fMRI alone, as well as enhance our understanding of BOLD effects.
• Studies of children at risk for disorders designed to identify predictors of psychopathology would be useful. Other "predictive" strategies would be prognostic.
• Resting perfusion studies may provide more stable, quantitative markers than conventional BOLD-fMRI cognitive paradigms. Perfusion studies may minimize variability across scanners and sites. Although temporal resolution is poorer, such studies may reflect low frequency activity and indicate connectivity.
• Studies directed toward better understanding of the baseline state or "default mode" and corresponding effects on the size or variance of the activation are needed.
• Pilot ASL (perfusion) projects should use simple fMRI tasks with resting or simple baselines across 3-4 sites in children, adolescents and young adults to determine if these produce biologically meaningful information.
• Simple, basic sensorimotor or psychophysiological tasks may help to anchor, track and index processes with potential cascading influences on higher-level processes.
• Independent components analysis (ICA) may be of value even with simple tasks, resting states or simple baselines such as visual fixation.

Technology, tools, and resources — needs

• There remains a need for improved understanding of fMRI signals, as well as the signature of other imaging modalities.
• There is a need for research to validate whether imaging measures and tools — e.g., cortical thickness (as seen on MRI) — are correlated with histological findings.
• Problems created by the lack of adequate tools may be compounded by researchers' misuse of these tools. Both may result in spurious findings when software is neither validated nor tested across sites in different situations.
• These needs are particularly pressing for pediatric neuroimaging, as standard image processing packages have been optimized for adult brains that are changing less rapidly than children's brains.
• Most major software packages have been developed at European institutions and have required ongoing support.
• Scientists in aging research might consider partnering with those in pediatric developmental research, as both fields share needs for tools to study age-related change.
• The field would benefit from further DTI statistical development, validation studies involving human and animal work in parallel, standard software packages, and methods for normalizing across age groups.
• Tools and a graphical user interface (GUI) are needed for integrating multiple imaging modalities (fMRI, aMRI, DTI, MRS) and to bring diverse datasets together into a spatially normalized template.