Metrics matter: Unlocking the secrets of brain networks and disease
June 3, 2026 — A team of University of Minnesota researchers have shed new light on the metrics scientists use to map the human brain, which could change the way we understand and treat neurological and psychiatric disorders, such as epilepsy, depression, and Parkinson’s disease.
Dysfunctional brain networks are at the root of these debilitating conditions. Visualizing how these networks interact is essential for developing effective, targeted therapies.
However, capturing a clear picture is incredibly challenging. The brain operates as a complex web of billions of interconnected neurons, and the accuracy of widely used mapping metrics under real-world data constraints has long been unclear.
In a new study published in the Journal of Neural Engineering, University of Minnesota researchers put these metrics to the test.
"Metric choice matters for reliably understanding brain networks, and brain networks matter for understanding neurological and psychiatric disease," says Kate Dembny, a co-author of the study and an MD/PhD candidate who’s pursuing her medical degree alongside a PhD in Biomedical Engineering through the University of Minnesota’s Medical Scientist Training Program.
The high stakes of picking the right metrics
Measures of network connectivity vary drastically in their ability to accurately reconstruct networks. The optimal metric changes based on the type of data collected and how much of the brain a researcher can see, known as network coverage. Choosing the wrong metrics can jeopardize the validity of the entire study.
"Picking a metric that can't accurately depict a brain network and using that reconstruction to try to understand a disease is like using a map of Chicago to find your way around Minneapolis," says Dembny.
Which metrics did best?
The research team found that metrics must be able to account for lags in time communication between brain regions to reliably reconstruct networks. Alarmingly, zero-lag communication measures, which are commonly used in the literature, performed no better than random chance.
Multivariate and bivariate transfer entropy proved to be extremely reliable. However, they require massive calculation times, limiting their practicality in today’s computing landscape.
Multivariate metrics excel at mapping large, sprawling networks. For smaller networks or when data about the network is more limited, bivariate metrics perform well.
Bivariate mutual information emerged as the most versatile tool, reliably reconstructing networks across the greatest number of contexts with reasonable computing times.
Conversely, the Granger Causality metric struggled to deliver accurate results across a wide variety of tested contexts.
A collaboration with the Medical School
Alongside Dembny, the research team includes Professor Tay Netoff and Hafsa Farooqi of the Department of Biomedical Engineering and Alexander Herman and David Darrow of the University of Minnesota Medical School.
Herman and Darrow are both practicing clinicians in addition to graduate faculty members with the Department of Biomedical Engineering. Together with Netoff, they form an integrative, interdisciplinary mentoring team for Dembny — one whose work exemplifies the cross-disciplinary collaboration essential to advancing understanding of neuropsychiatric disease and developing novel neuromodulatory therapies to care for patients.
This research was funded by the Data Science Initiative-MnDRIVE PhD Graduate Assistantship program, the National Institute of Drug Addiction (K23DA050909), the Brain & Behavior Research Foundation (NARSAD Young Investigator Award #28426), the University of Minnesota’s MnDRIVE (Minnesota’s Discovery Research and Innovation Economy) Initiative, and the University of Minnesota Medical Discovery Team on Addiction.