mental illness

Brain structures and mental disorders: what’s the connection?

Computer science doctoral student Bohan Xu (MS ’16) wants to understand more thoroughly brain-structure differences between individuals with psychiatric disorders and people in a “healthy” control group. Xu is undertaking this study at the Laureate Institute for Brain Research (LIBR) under the supervision of Rayus Kuplicki (BS ’09, MS ’11, Ph.D. ’14) from LIBR and Professor Sandip Sen. He is joined on the project by a number of other TU students: Mahdi Moradi, who is completing a Ph.D. in computer science; as well as Kelly Cosgrove, Danielle Deville, McKenna Pierson and Timothy McDermott, all of whom are clinical psychology doctoral candidates.

Computer science student Bohan Xu wearing glasses and a white open-collar shirt
Bohan Xu

“Unlike diabetes or cancer, which can be diagnosed by medical tests,” remarked Xu, “most mental illnesses are determined by the psychological evaluation a physician or mental health professional carries out when they talk to you about your symptoms, thoughts, feelings and behavior patterns. Questionnaires are also sometimes used to help gather pertinent information. One goal at LIBR, however, is to find out a more reliable and precise way to identify mental disorders.”

Peering within the brain for answers

For the last few years as a TU/LIBR researcher, Xu’s work has focused on data analysis within the broad field of psychiatry; for example, the potential for diagnosing mental illness based on the concentrations of volatile organic compounds in exhaled breath and the relations between depression and C-reactive protein. Like his current undertaking, those projects entailed studying the relationship between subjects’ mental disorders/health status and different variables about their demographic information, bioassay tests and brain images.

Rayus Kuplicki smiling and wearing an open-collar blue shirt
Rayus Kuplicki

At present, Xu and his fellow investigators are using brain-structure image data (voxel-wise gray matter volume) from healthy controls to build a normative regression model that accounts for age and gender. This trained normative model then estimates a normal range of gray matter volume based on the healthy controls.

“The more that patients deviate from those healthy controls, the more likely their observed gray matter volumes will be outside the estimated ranges,” Xu explained. “Furthermore, we should be able to use these deviations to locate the potential areas of the brain that result in mental disorders.”

To date, Xu and his team have built the normative model. When verifying that model, however, they discovered something they could not readily understand. “The healthy controls are randomly divided into two groups,” Xu explained. “First, one group is used to build the normative model and the other group is used to validate the model; however, we observed a constant pattern of deviation when validating the model, which is not supposed to happen since the validation group are healthy controls as well.” The team’s present task, therefore, entails searching for an answer to that anomaly.


Learn more about the graduate studies pathways available through TU’s Tandy School of Computer Science.