This article is more than 5 years old. Information may no longer be current.
Machine learning links brain connectivity patterns with psychiatric symptoms
Click Here to Manage Email Alerts
Using machine learning, researchers from the University of Pennsylvania found that four dimensions of psychopathology — mood, psychosis, fear and disruptive externalizing behavior — were linked to distinct patterns of connectivity in the brain.
“Unlike other branches of modern medicine which use biologically-based tests of disease, psychiatry still relies on patient reports and physician observations for clinical decision-making instead,” Theodore D. Satterthwaite, MD, assistant professor, department of psychiatry, Perelman School of Medicine, told Healio Psychiatry.
“One obstacle for developing biomarkers for psychiatry is that there is not a clear mapping between abnormalities in the brain and psychiatric disorders,” he continued. “This lack of correspondence is evident in both the high levels of comorbidity we see among disorders (eg, depression often goes with anxiety) and that the heterogeneity within disorders (eg, depression presents differently among people). One potential candidate of biomarker for psychiatry is brain networks, which are brain regions that are densely connected to each other, and work together in cognitive processes such as self-control, memory, and motor function.”
Using a machine learning technique known as sparse canonical correlation analysis, the investigators sought to uncover the brain networks tied to psychiatric disorders in a sample of 999 youth aged 8 to 22 years. Participants completed both functional MRI scans and a comprehensive assessment of psychiatric symptoms as part of the Philadelphia Neurodevelopmental Cohort, according to the press release.
Analysis showed patterns of changes in brain networks that were related to psychiatric symptoms. Specifically, the researchers found four distinct dimensions of psychopathology — mood, psychosis, fear and disruptive behavior — were linked to a distinct pattern of irregular connectivity across the brain. Loss of network separation between the default mode network and executive networks were shared across all dimensions, the researchers found.
“Each of these dimensions was predicted by unique abnormalities of brain networks,” Satterthwaire explained. “However, all dimensions were marked by abnormally high levels of connectivity between the default mode network and fronto-parietal network, two brain regions that usually become increasingly distinct as the brain matures. This loss of normal brain network segregation supports the hypothesis that many psychiatric illnesses may be disorders of brain development.”
According to the press release, the findings demonstrated each dimension contained symptoms from different clinical diagnostic categories — for example, the mood dimension comprised symptoms from depression, mania and obsessive-compulsive disorder. Although the disruptive externalizing behavior dimension comprised symptoms from ADHD and oppositional defiant disorder, it also included irritability from the depression domain.
Furthermore, the authors found that connectivity associated with mood and psychosis became more prominent with age, and sex differences were found in connectivity associated with mood and fear. According to the release, brain connectivity patterns tied to mood and fear were stronger in females than males.
“This study demonstrates the importance of using diverse data types to study mechanisms of psychopathology across clinical diagnostic boundaries,” Satterthwaite said. “Moving forward, we hope to integrate genomic data in order to describe pathways from genes to brain to symptoms, which could ultimately be the basis for novel treatments for mental illness.”
“Future breakthroughs in brain science to understand mental illness require large amount of data,” first author Cedric Xia, an MD-PhD candidate at the Perelman School of Medicine, told Healio Psychiatry. “While the current study takes advantage of one of the largest samples of youth, the size remains dwarfed by the complexity of the brain. The neuroscience community is actively working towards collecting higher quality data in even larger samples, so we can validate and build upon the findings.” – by Savannah Demko
Disclosure: Satterthwaite reports no relevant financial disclosures. Please see the study for all other authors’ relevant financial disclosures.
Perspective
Back to Top
Joshua L. Roffman, MD, MMSc
A uniquely challenging aspect of psychiatry is its lack of clinically meaningful biological markers to guide diagnosis, prognosis and intervention. To address the persistent gap between neuroscientific and clinical innovation, NIMH developed the Research Domain Criteria (RDoC) framework, and has prioritized research adopting this approach. Fundamentally, RDoC aims to break down mental illness into its genetic, molecular and circuit-level building blocks, and to link these components with patterns of symptoms and behavior that are parsed by their underlying biology, rather than by traditional diagnostic criteria.
The hope is that this bottom-up approach will ultimately deliver more tailored and efficient care, and the elegant work of Cedric Xia, Theodore Satterthwaite and colleagues takes a meaningful step in this direction. Adolescents are an especially important group to study as they approach the age of greatest risk for full-blown illness, and yet clinical predictors of such progression are often subtle and nonspecific. Using an unbiased, data-driven method, the investigators have begun to disambiguate common and syndrome-specific patterns of altered brain connectivity in youths with sub-diagnostic psychiatric symptoms. For example, while impaired differentiation of the brain’s default network was seen as a common element across psychopathology, the specific networks that failed to segregate from the default network varied by symptom group. Importantly, the investigators also demonstrate that network connectivity is a moving target throughout adolescence, a reminder that preventive interventions may have limited launch windows with respect to brain development.
There remain many challenges to overcome before the RDoC approach translates into improved patient care. For one, like all brain imaging research in psychiatry, the work of Xia and colleagues relies on group-level data to link brain phenotypes with behavioral constructs. For biomarkers to have clinical utility, they must provide valid and predictive individual-level data, which for imaging data will ultimately require some combination of stronger effect size and better in vivo resolution. An additional obstacle is linking imaging data with individual-level indices of genetic risk, which remains the strongest predisposing factor for most major psychiatric conditions. Recent findings from the Brainstorm Consortium vividly demonstrate the byzantine entanglement of genomic risk among DSM-defined psychiatric disorders, and the need to better define the biological pathways that lead from genes to specific pathology.
Just as the brain does not respect conventional diagnostic boundaries, it will take large-scale collaborative efforts from investigators who eschew academic silos to tackle these challenges and develop brain-based treatments for those seeking psychiatric care. Efforts like this one — merging expertise in imaging, epidemiology, neuropsychology, and big data analytics — lead by example.
Click Here to Manage Email Alerts