Summary: A research team developed a machine learning model that predicts individual tendencies to engage in persistent negative thinking, known as rumination, from patterns of brain connectivity measured with functional MRI (fMRI).
Rumination—repetitive, self-focused thinking about negative feelings, past mistakes, or worries—is a known risk factor for depression and anxiety. Identifying objective neural markers of rumination could improve early detection, guide treatment decisions, and provide measurable endpoints for tracking clinical progress.
Key findings
- Machine learning models were successfully trained to estimate self-reported rumination scores from resting-state fMRI data by using measures of dynamic functional connectivity.
- Among regions of the brain’s Default Mode Network (DMN), only models focusing on the dorsomedial prefrontal cortex (dmPFC) robustly predicted rumination scores in healthy participants.
- The dmPFC-based model also generalized to clinical data: it predicted depression symptom scores in patients with Major Depressive Disorder (MDD), suggesting potential utility as a biomarker.
Led by KIM Jungwoo at the Center for Neuroscience Imaging Research (CNIR), Institute for Basic Science (IBS), in collaboration with researchers from the University of Arizona and Dartmouth College, the study used a predictive modeling approach to isolate a neural signature of rumination. Earlier work had implicated the Default Mode Network in rumination, but which DMN nodes best explain individual differences remained unresolved.
The researchers focused on the temporal stability of connectivity—specifically, the variance of dynamic connectivity between each DMN node and the rest of the brain. Because rumination is inherently repetitive and temporally persistent, the team hypothesized that higher variance in time-varying connectivity patterns might reflect individual tendencies to engage in sustained negative thought.
To evaluate this, the investigators recorded resting-state fMRI from healthy volunteers and computed the variance of dynamic functional connectivity between each DMN region and the whole brain. These connectivity-variance measures served as input features for machine learning models trained to predict participants’ self-reported rumination scores.
Across the DMN, models centered on the dmPFC emerged as the most predictive. Further analysis identified dynamic connectivity between the dmPFC and specific regions—notably the inferior frontal gyrus and parts of the cerebellum—as particularly informative for estimating rumination severity. These connections likely reflect interactions between self-referential, cognitive-control, and affective processing systems that support repetitive thought.
Importantly, the dmPFC-based dynamic connectivity marker was not only predictive in subclinical samples but also generalized to clinical cases. A refined marker derived from the strongest predictive features successfully predicted depression scores in a sample of adults diagnosed with Major Depressive Disorder. This cross-sample generalizability supports the marker’s potential as a biomarker for risk and symptom severity in depression.
The findings bolster an emerging view that the dmPFC plays a central role in trait rumination and in the cognitive processes that can precipitate or maintain depressive symptoms. By linking time-varying patterns of connectivity to individual differences in repetitive negative thinking, the study provides a promising avenue for objective assessment of rumination using neuroimaging combined with machine learning.
Professor WOO Choong-Wan, a lead author on the study, emphasized that natural fluctuations in thought influence mood and emotion. He noted that the tendency to ruminate can be decoded from fMRI connectivity patterns and expressed the hope that neuroimaging measures like this may one day support monitoring and management of mental health.
Next steps planned by the team include validating and refining the predictive model in larger, more diverse populations and exploring clinical applications. Integrating such neuroimaging markers with existing diagnostic and therapeutic strategies could enable more personalized interventions that directly target rumination and its role in depression.
About this machine learning and rumination research news
Author: William Suh
Source: Institute for Basic Science
Contact: William Suh – Institute for Basic Science
Image: The image is credited to Neuroscience News
Original Research: Open access. “A dorsomedial prefrontal cortex-based dynamic functional connectivity model of rumination” by KIM Jungwoo et al., Nature Communications
Abstract
A dorsomedial prefrontal cortex-based dynamic functional connectivity model of rumination
Rumination is a pattern of repetitive thinking about negative internal states and is commonly associated with depression. Although prior studies have linked trait rumination to alterations in Default Mode Network function, predictive neuroimaging markers have been limited. This study uses the variance of dynamic resting-state functional connectivity to build a predictive marker of rumination and tests that marker across five diverse subclinical and clinical samples (total n = 288).
A whole-brain marker based on dynamic connectivity with the dmPFC generalized across subclinical datasets, and a refined marker composed of the most informative features further predicted depression scores in adults with Major Depressive Disorder (n = 35). These results highlight the role of the dmPFC in trait rumination and offer a dynamic functional connectivity marker that may be useful for research and clinical applications related to rumination and depression.