Diagnostic and predictive value of structural and high-order functional brain topology in major depressive disorder.
Authors
Affiliations (6)
Affiliations (6)
- Department of Radiology, Yantai Affiliated Hospital of Shandong Medical and Pharmaceutical University, Yantai, Shandong, China.
- School of Medical Imaging, Shandong Medical and Pharmaceutical University, Yantai, Shandong, China.
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China.
- Shandong Provincial Key Medical and Health Laboratory of Intelligent Diagnosis and Treatment for Women's Diseases (Yantai Yuhuangding Hospital), Yantai, China.
- Big Data and Artificial Intelligence Laboratory, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China.
- Shandong Provincial Key Laboratory of Neuroimmune Interaction and Regulation, Yantai, Shandong, China.
Abstract
Conventional neuroimaging analyses for major depressive disorder (MDD) predominantly construct brain networks based solely on pairwise regional functional couplings. Accumulating microscale neuroscientific evidence highlights the essential role of higher-order interactions (HOIs, interactions among three or more brain nodes) in capturing the intricate spatiotemporal dynamics of the human brain. This study aims to integrate machine learning techniques with neuroimaging topological features to systematically investigate the impact of non-pairwise interaction patterns in the brains of MDD patients on system dynamics, and to develop diagnostic and predictive models for MDD clinical manifestations. Firstly, diffusion tensor imaging (DTI) and resting-state functional magnetic resonance imaging (rs-fMRI) data were acquired from 154 MDD patients with first-episode, drug-naïve and 153 age and gender matched healthy controls (HCs). Graph-theoretical analysis was applied to DTI-derived tractography data to quantify structural topology metrics, whereas persistent homology was adopted to extract higher-order functional topological signatures from rs-fMRI time series. Second, feature-level fusion was performed, and then a classification model was constructed using support vector machine (SVM) for MDD diagnosis. A repeated nested validation strategy with 50 repeated stratified 70/30 hold-out splits was adopted. Classification performance was evaluated using the area under the receiver operating characteristic curve (AUC). Subsequently, a prediction model was built based on the selected fused features using ridge regression, and the prediction performance was evaluated based on the correlation between the predicted values and the actual scale scores. Compared with HC, MDD patients exhibited significant alterations in structural and functional topological metrics within key brain networks, including DMN, VN and FPN. The classification performance of the SVM model integrating multimodal neuroimaging features achieved excellent classification performance, with an AUC of 88.2%. Additionally, multimodal imaging features demonstrated robust predictive capacity for suicide risk level in MDD patients (r=0.604). By integrating structural and higher-order functional topological features, we are able to effectively diagnose and predict MDD, highlighting the crucial role of higher-order interactions in the pathological mechanisms of brain networks in depression.