Understanding the Value and Limitations of MRI Neuroimaging in Psychiatry

In psychiatry, MRI neuroimaging methods have become a critical tool for performing research with the intent of developing a better understanding of brain circuit alterations that are associated with etiology, pathophysiology, and treatment response. It is hoped that such an understanding will eventually lead to the capacity to better identify 1) the factors associated with the early-life risk for illness, 2) the brain systems that underlie the heterogeneity in symptom presentations within and across illnesses, 3) the brain changes related to treatment response, and 4) new ideas for developing illness-specific neural treatment targets. However, there are significant concerns that must be considered when making assertions about psychiatric illnesses from neuroimaging data.

While we can now view detailed, impressive images of the brain constructed from MRIs, estimate regional brain responsivity to specific tasks, and gauge their activity at “rest,” there are important caveats to keep in mind when interpreting imaging data. For example, MRI only provides a macro view of the brain’s structure and inner workings. The resolution of 3-T MRI for structural images can range but is typically a volume of about 1 × 1×1 mm³. For functional imaging, the resolution is typically around 2 × 2×2 mm³. To put the resolution of MRI in perspective, a recent report from a study using serial section electron microscopy estimated that 1 mm³ of human brain tissue contains 50,000 cells and 130 million synaptic connections (1).

As pointed out by Weinberger and Radulescu (2) in a previous commentary discussing the use of MRI in psychiatric research, the images generated from MRI data are only indirectly related to brain structure and function. For example, structural measures of gray and white matter reflect the magnetic resonance properties of hydrogen protons in tissues with different water and fat concentrations. Diffusion tensor imaging is a method that allows for the ability to image myelinated white matter pathways. This method is performed by constructing images from MRI sequences that assess the direction of water diffusion in the brain such that diffusion parallel to axon tracts is more rapid and moves in a unified direction relative to the water diffusion that moves perpendicularly to these tracts. Task-based functional MRI capitalizes on assessing the blood-oxygen-level-dependent (BOLD) signal, which gauges regional differences in blood oxygenation levels and cerebral blood flow that are assumed to reflect local regional brain activity. Finally, acquiring measures of the BOLD signal at “rest” and then temporally correlating these signals across different brain regions is considered to be a measure of resting-state functional coupling. It is important to understand that numerous factors unrelated to the inherent state of the brain can influence these indirect measures of brain structure and function acquired with MRI. For example, excessive movement in the scanner can result in changes in signal intensity and, if not accounted for, can lead to the misinterpretation of results relevant to brain structure, task-related activity, or resting-state functional coupling.

Even with these concerns, there is no question that neuroimaging methods are critical for deepening the understanding of the structure and function of the human brain. In this regard, MRI has become a standard tool for investigating brain alterations that are associated with psychiatric illnesses. In a reflection on the history and future use of MRI in psychiatry, Etkin commented on the lack of reproducible, diagnosis-specific neuroimaging findings (3). This point is consistent with the transdiagnostic nature of many neuroimaging findings and also draws attention to concerns related to our current diagnostic system. Etkin also discussed the relatively weak effect sizes associated with findings from very large cross-sectional neuroimaging studies and suggested an alternative approach—within-subject repeated longitudinal imaging—that may be more relevant to advancing ideas related to the development of individualized treatment strategies. It is important to emphasize that, in general, neuroimaging data are associational, not allowing for an examination of causal mechanisms. This issue can, in part, be overcome by designing experiments that use neuroimaging to assess the effects of experimentally perturbed brain function. This can be accomplished by using imaging to assess the effects of interventions such as transcranial magnetic stimulation aimed at modulating regional brain activity or the administration of pharmacological agents directed at affecting neurochemical systems. Thus, questions can be addressed that are related to the role of a brain region or neurochemical system in mediating the MRI findings that are linked to a psychological or cognitive process. Finally, it is important to emphasize that translational cross-species studies are critical in providing a deeper understanding of mechanisms. The capacity to selectively modify the function of specific neurons or circuits in the appropriate animal model with methods such as optogenetics or chemogenetics is an important resource for elucidating the mechanisms that underlie findings from human neuroimaging studies.

In this issue of the Journal, we present a series of papers that highlight the use of various neuroimaging techniques that are aimed at linking brain structure and function to 1) understanding pathophysiological processes associated with illnesses, 2) factors related to normative development and early-life risk, and 3) treatment response. In this regard, we begin with an overview on MRI findings in ADHD authored by Dr. Xavier Castellanos from New York University in which he emphasizes the importance of large-scale imaging/phenotyping studies, the within-subject reliability of structural MRI, and the transdiagnostic nature of many of the ADHD-related imaging findings (4). We also include two other reviews in this issue. In the first, Pine et al. (5) consider an evolutionary perspective, from nonhuman primate to human brain, in discussing brain systems that underlie the development of cognitive and emotion-related capacities relevant to developing stress-related psychopathology. The second review, by Grzenda and colleagues (6), is a primer on the basic principles of machine learning as they relate to psychiatry. It is clear that applications of artificial intelligence will become increasingly important in the future, enabling new opportunities to integrate large, complex multimodal data sets in the service of making predictions about prognosis and treatment outcomes at the level of the individual patient.

White Matter Alterations Are Generally Shared Across ADHD and Autism

White matter makes up a large portion of the brain and is critical in providing structural connectivity across brain regions, and the myelin sheath surrounding axons modulates the fine-tuning of the timing of information transfer across communicating brain regions. Using diffusion-weighted MRI, various white matter alterations have been reported across different disorders. Tung et al. (7) assessed the extent to which white matter parameters in a large cohort of individuals with ADHD (N=279, with 121 unaffected siblings) and autism spectrum disorder (ASD) (N=175, with 72 unaffected siblings) deviated from white matter in typically developing subjects (N=626). The authors were motivated to understand the similarities and differences in white matter alterations between ADHD and ASD because these illnesses share some symptomatic features, can have overlapping presentations, and are neurodevelopmental in origin. While some differences were found between the disorders, relative to the comparison subjects, alterations in the estimated developmental trajectory of various white matter tracts were shared across both patient groups. Involved tracts broadly included the corpus callosum, some cortical-cortical tracts, and some cortical-subcortical tracts. The authors also found that unaffected siblings of patients with ADHD and ASD had white matter profiles that were related to their affected sib. Additionally, within the patient groups, there was more individual variation in the white matter parameters of the altered tracts, termed an increase in “idiosyncrasy.” The authors interpreted this finding as being possibly related to the wide range of heterogeneity of symptoms seen in ADHD and ASD. Also, along these lines, the authors found that different cognitive and psychopathological features correlated with alterations in specific white matter tracts. Taken together, these alterations in white matter tracts, shared across ADHD and ASD patients, are consistent with numerous other neuroimaging findings that appear to be transdiagnostic in nature and not illness specific.

Resting-State Functional Connectivity Is Associated With Treatment Response in ADHD

Norman et al. (8) aimed to understand the longitudinal, within-individual relationship between resting-state functional coupling MRI measures and ADHD symptoms in children and adolescents treated with stimulants. Based on previous studies related to ADHD, the focus of this work was on the cingulo-opercular, striatal-thalamic, and default-mode networks. Study subjects (110 with ADHD and 142 typically developing comparison youths) were followed longitudinally in relation to their naturalistic treatment with stimulants and were scanned up to five times with resting-state MRI. The findings revealed that worse responses to treatment were associated with an age-related atypical increase in connectivity in the cingulo-opercular network. Better responses to treatment were associated with appropriate age-related cingulo-opercular connectivity. The frontal operculum, insula, dorsal anterior cingulate cortex, and posterior medial prefrontal cortex are components of the cingulo-opercular network, and it was previously shown in children with ADHD that acute stimulant administration was associated with the normalization of the function of this network. In general, this finding is consistent with other studies in other psychiatric illnesses and suggests that individuals with more intact functioning of hypothetically involved circuits are more likely to be responsive to treatments. In an editorial provided by Dr. Samuele Cortese, from the University of Southampton in the U.K., the findings from this study are reviewed and contextualized in relation to biomarker development and the steps necessary to validate algorithms for personalized medicine approaches (9). He concludes that the findings from the Norman et al. study are an important but very early step in this process.

Developmental Trajectories of Social Communication Deficits as They Relate to Autism Spectrum Disorder

In an attempt to better understand how developmental trajectories of altered social communication relate to ASD, Riglin and colleagues present a paper that reports on longitudinal data from a large set of unselected individuals (N=8,094) assessed from ages 7 to 25 (10). By using the Social and Communication Disorders Checklist (SCDC), the authors identified three different developmental trajectories over this time period. Eighty-eight percent of participants were characterized to be in the low SCDC score trajectory group, whereas 5% were in the declining SCDC score group, and 6.5% were in the late-emerging group. To assess ASD, childhood diagnoses were used and “risk” for ASD diagnosis in adulthood was assessed with the Autism Spectrum Quotient, completed by the young adult and a parent. Childhood diagnoses of ASD were very low in the low SCDC and late-emerging trajectory groups (<5%) and were around 9% in the declining-trajectory SCDC group. When using the parent rating for ASD risk that was assessed during early adulthood, high risk for adult ASD was very low for the low-trajectory group, around 50% for the late-emerging group, and less than 20% for the declining-trajectory group. The finding that ASD symptoms decreased with age in the declining-trajectory group, the group with the highest levels of childhood autism diagnoses, is consistent with previous reports. In relation to the late-emerging trajectory group, the authors interpret the data to support the existence of a developmental trajectory for ASD that is characterized by the late emergence of symptoms and disability. In their editorial on this paper, Dr. Catherine Lord, from UCLA, and Dr. Somer Bishop, from UCSF, review the findings and highlight the design and value of the study in tracking social deficits across development (11). They also challenge a variety of assumptions made in the paper, including the notion that ASD, as an illness, is on the extreme end of a continuum of “autistic traits.” Further, Lord and Bishop emphasize the importance of not equating deficits in social behavior with ASD, as social deficits are nonspecifically related to a variety of developmental problems.

Transdiagnostic Alterations in Mid-Insula Activation During Interoception

It has been suggested that various psychiatric disorders are associated with alterations in interoceptive processes, and the paper by Nord et al. in this issue presents findings from a neuroimaging meta-analysis characterizing the transdiagnostic neural correlates of interoception (12). Interoceptive processes are critical for survival, as they support the maintenance of homeostasis, in part by providing individuals with information about their internal physiological state. In this regard, the insular cortex has been identified to be critically involved in awareness and in particular in the interoceptive processing of bodily states. The insula can be subdivided into anterior, mid, and posterior portions, with the anterior insula more related to cognitive, emotional, and affective processing, the posterior insula more related to somatosensory processing, and the mid-insula providing connectivity between the anterior and posterior portions. By comparing fMRI data from 626 patients with a range of psychiatric diagnoses to 610 control subjects, the authors found altered interoception-related left dorsal mid-insula activation to be common across the patient groups (diagnoses included schizophrenia, bipolar disorder, depression, anxiety disorders, posttraumatic stress disorder, eating disorders, and substance use disorders). The authors further demonstrated that the identified mid-insula region did not overlap with regions that are associated with affective processing or regions that are associated with interventions for affective disorders. In their editorial, Drs. Martin Paulus and Sahib Khalsa, from the Laureate Institute for Brain Research, discuss the implications of this finding, highlighting the connectivity of this relatively underappreciated mid-insular cortex region to the midcingulate cortex, a region involved in cognitive control and decision making (13). They also discuss “interoception prediction errors” and their own work in relation to how they understand the development of interoception-related psychopathology.

Using fMRI in Newborns to Understand Neural Systems Associated With the Risk for Developing Pathological Anxiety

Sylvester and colleagues use fMRI in sleeping neonates to examine the neural systems that are associated with infants’ responses to the presentation of unexpected auditory stimuli (14). Such a paradigm probes the early and fundamental automatic responses to novelty, which later in life are highly relevant to the assessment of, responding to, and coping with potential threat. Furthermore, increased responsivity to unexpected stimuli early in life has been associated with the development of pathological anxiety, and studies in adults have characterized the neural correlates related to the presentation of unexpected stimuli. In imaging 41 infants from a normative sample (average age, 28 days), the researchers characterized the infants’ responses to deviant sounds, identifying regions of activation that are similar to those observed in adults. Some of these regions included the thalamus, dorsal anterior cingulate cortex (ACC), anterior insular cortex, and dorsal medial frontal cortex. Additionally, the researchers examined the extent to which individual differences in levels of maternal trait anxiety were related to their infant’s pattern of brain activation. Results demonstrated numerous brain regions in the neonates in which brain activation was significantly correlated with maternal trait anxiety, including the anterior insula, dorsal ACC, subgenual ACC, and ventrolateral prefrontal cortex. Notably, altered activation of these regions has also been associated with anxiety disorders. The authors point out that high levels of maternal anxiety confer significant risk for offspring to develop anxiety disorders and suggest that the early-life neural findings uncovered by their work may represent the earliest brain manifestations of this risk.

Conclusions

The papers presented in this issue of the Journal highlight the use of MRI as a tool to understand psychopathology. The findings from these papers reveal structural and functional brain alterations that are associated with illnesses such as ADHD, ASD, and anxiety disorders. The findings include the observation that similar white matter alterations are shared across ADHD and ASD and that functional connectivity of the cingulo-opercular network is associated with treatment response in ADHD patients. Other papers in this issue characterize neural circuits that are associated with processes that are related to the diverse presentations of psychopathology. These include identifying altered interoception-related activation of the mid-insula as being common across psychiatric diagnoses and identifying, in neonates, brain regions that are activated by the presentation of unexpected stimuli that are correlated with maternal anxiety. Despite these potentially important findings, which lend new insights into brain alterations associated with psychopathology, the extent to which they are related to etiology or pathophysiology remains to be determined. By their nature, the findings cannot address questions of causality. After years of neuroimaging research, it is apparent that there are considerable limitations to current imaging methods that challenge the earlier enthusiasm about the use of MRI in psychiatry. At the same time, it is important to recognize the very significant contributions from neuroimaging studies that have allowed for an in-depth understanding of brain-behavior relationships in humans. These studies, like those presented in this issue, provide new ideas about adaptive and maladaptive cognitive, emotional, and behavioral processes that are relevant to psychiatric illnesses. Neuroimaging measures and other measures of brain activity such as EEG are already being explored in the development of personalized treatment approaches. In pursuit of this goal, it will be critical to take advantage of machine learning methods in relation to large-scale neuroimaging, genomic, and demographic data that can be used to develop algorithms that are reliable at the individual patient level.