Individualized prediction of multi-domain intelligence quotient in bipolar disorder patients using resting-state functional connectivity

Background

Although accumulating studies have explored the neural underpinnings of intelligence quotient (IQ) in patients with bipolar disorder (BD), these studies utilized a classification/comparison scheme that emphasized differences between BD and healthy controls at a group level. The present study aimed to infer BD patients’ IQ scores at the individual level using a prediction model.

Methods

We applied a cross-validated Connectome-based Predictive Modeling (CPM) framework using resting-state fMRI functional connectivity (FCs) to predict BD patients’ IQ scores, including verbal IQ (VIQ), performance IQ (PIQ), and full-scale IQ (FSIQ). For each IQ domain, we selected the FCs that contributed to the predictions and described their distribution across eight widely-recognized functional networks. Moreover, we further explored the overlapping patterns of the contributed FCs for different IQ domains.

Results

The CPM achieved statistically significant prediction performance for three IQ domains in BD patients. Regarding the contributed FCs, we observed a widespread distribution of internetwork FCs across somatomotor, visual, dorsal attention, and ventral attention networks, demonstrating their correspondence with aberrant FCs correlated to cognition deficits in BD patients. A convergent pattern in terms of contributed FCs for different IQ domains was observed, as evidenced by the shared-FCs with a leftward hemispheric dominance.

Conclusions

The present study preliminarily explored the feasibility of inferring individual IQ scores in BD patients using the FCs-based CPM framework. It is a step toward the development of applicable techniques for quantitative and objective cognitive assessment in BD patients and contributes novel insights into understanding the complex neural mechanisms underlying different IQ domains.