Disrupted brain connectivity in postpartum depression: Insights from resting-state fMRI and machine learning
Can Liu1, Tong Zhu2, Mingke Liu3
1Department of Radiology, The Affiliated Yongchuan Hospital of Chongqing Medical University, Chongqing 402160, China; College of Artificial Intelligence Medicine, Chongqing Medical University, Chongqing 400016, China.
Background:
Postpartum depression (PPD) is a common women's psychological health issue. While studies have identified regional functional abnormalities, the global functional topological alterations associated with PPD remain to be fully characterized. This study aims to investigate the alteration of functional topological properties in PPD patients.
Methods:
Resting-state functional MRI (rs-fMRI) was acquired from 30 PPD patients, 23 healthy pregnant women (HPW), and 26 healthy non-pregnant women (HC). Functional brain networks were constructed using inter-regional Pearson's correlation coefficient and analyzed via graph theory. Machine learning was applied to the functional connectome to distinguish PPD from HPW.
Results:
Compared to HC and HPW, the PPD group showed a shift toward a more regularized network topology in functional brain network. In comparison with HC, PPD had altered topological properties mainly in the salience network (SN, e.g., left insula) and associated subcortical regions (e.g., amygdala), while HPW exhibited functional differences mainly within the default mode network (DMN). Abnormal regions (e.g., pallidum, precuneus) between PPD and HPW correlated with depression severity. Combining machine learning with functional connectivity metrics predicted PPD with 88 % accuracy.
Conclusion:
Pregnancy may alter the functional connectome in DMN, and postpartum depression may disrupt the connectivity in SN. The insula and precuneus are critical for identifying PPD and HPW. These findings suggest that functional connectome alterations are clinical significant and may facilitate the timely clinical detection of PPD.
