Comparative Evaluation of Stress-Induced Brain Connectivity Changes in Pre-Eclamptic Versus Normotensive Women: Harnessing AI for Advanced Neuroimaging Data Analytics

Abstract

Background: Artificial intelligence (AI) offers new opportunities for analyzing high-dimensional neuroimaging data to uncover subtle stress-related brain connectivity changes. This study applied AI-driven methods to compare stress-induced brain connectivity in pre-eclamptic versus normotensive women.

Aim: To evaluate stress-induced brain connectivity changes in pre-eclamptic versus normotensive women using AI-based neuroimaging data analytics.

Methods: A prospective comparative study was conducted on 80 participants (40 pre-eclamptic, 40 normotensive) recruited at a tertiary care center. Stress-induction paradigms were performed during fMRI acquisition, supplemented with diffusion tensor imaging. Connectivity matrices were analyzed using convolutional neural networks, graph neural networks, support vector machines, and autoencoders. Graph-theoretic measures (efficiency, modularity, clustering, hub disruption) were calculated. Model performance was validated using cross-validation, with interpretability provided through SHAP and attention mechanisms.

Results: Pre-eclamptic women exhibited significantly greater dynamic functional connectivity variance (0.87 ± 0.19 vs. 0.73 ± 0.17; p=0.001), higher temporal switching rates (3.90 ± 1.10 vs. 3.20 ± 0.90; p=0.003), and reduced network modularity (0.41 ± 0.07 vs. 0.45 ± 0.06; p=0.005). CNNs and GNNs identified more altered edges and higher anomaly scores in pre-eclampsia. AI-based biomarkers showed strong predictive accuracy with cross-validated AUC (0.88 vs. 0.82; p<0.001) and F1-score (0.81 vs. 0.72; p<0.001). SHAP analysis highlighted limbic edges as key discriminators.

Conclusion: AI models reliably identified stress-related brain network alterations in pre-eclamptic women, outperforming traditional measures. These findings validate the potential of AI-derived biomarkers for precision neuroimaging and early risk stratification in maternal brain health.