Chongqing Medical University Team: Dual-Branch Graph Attention Network Enables Personalized Prediction of ECT Efficacy in Adolescent Depression
Published 02 February, 2026
Major depressive disorder (MDD) poses a global health challenge, particularly among adolescents, where it is characterized by high recurrence and suicide rates. While electroconvulsive therapy (ECT) is effective for rapidly alleviating symptoms, its usage in adolescents is limited, partly due to the inability to predict individual treatment outcomes accurately.
Traditional prediction methods often rely on single-modality imaging or manual feature selection, which fails to capture the complex, non-linear network architecture of the brain.T o overcome these limitations, a research team from the College of Medical Informatics at Chongqing Medical University and the Department of Radiology at The First Affiliated Hospital of Chongqing Medical University, developed a "Dual-Branch Graph Attention Network" (DBGAN) which integrates structural MRI (sMRI) and functional MRI (fMRI) data within a unified system.
The study, published in the KeAi journal Meta-Radiology, represents a promising step toward precision medicine in psychiatry, offering a robust framework for handling high-dimensional, heterogeneous brain network data even with limited sample sizes.
“In our study involving 27 adolescent MDD patients, the DBGAN model achieved a mean accuracy of 85.3% and an F1-score of 0.905 in predicting treatment responders, significantly outperforming traditional machine learning models (such as SVM and Random Forest) and standard Convolutional Neural Networks (CNNs),” shares corresponding author Du Lei. “Our study highlights that relying on a single imaging modality may overlook complementary information," the authors explain. "By integrating brain network data from both structural and functional MRI, we can capture coordinated changes that are critical for identifying responders.”
Notably, the model's decision-making was driven by distinct brain regions: functional signals from the right posterior insula and dorsal cingulate gyrus, and structural signals from limbic areas like the left amygdala and right hippocampus. “These regions are closely linked to emotion regulation and memory processing, providing neurobiological validation for the model's predictions,” says Lei. “This approach supports the efficient allocation of medical resources and helps prevent unnecessary interventions for patients unlikely to respond.”
Contact author:
Du Lei, PhD, College of Medical Informatics, Chongqing Medical University, alien18@163.com
Funder:
This work was supported by Chongqing Talents Exceptional Young Talents Project (cstc2024ycjh-bgzxm0220 (DL)) and Natural Science Foundation of Chongqing (CSTB2024NSCQ-MSX0377 (DL)).
Conflict of interest:
The authors declare that they have no known competing financia.interests or personal relationships that could have appeared to influencethe work reported in this paper.
See the article:
Jingyu Zhang, et al. "Predicting Efficacy of Electroconvulsive Therapy for Adolescent Major Depressive Disorder Using a Dual-Branch Graph Attention Network Fusing Multi-modal MRI." Meta-Radiology 3.4 (2025): 100184.