Background: Resting-state functional MRI (rs-fMRI) studies typically rely on linear Pearson correlation to characterize brain connectivity, potentially overlooking the distributional characteristics of functional networks. This study introduces a kernel-transformed functional connectivity (FC) entropy framework to quantify network dedifferentiation in bipolar disorder (BD). Methods: We utilized a Gaussian kernel function to execute a nonlinear similarity transformation (referred to as reweighting) on standard linear correlation matrices. This approach acts as a functional filter to amplify the contrast between strong and weak connections. Multiscale entropy (global, modular, and nodal) was subsequently calculated to characterize the uniformity of connectivity weight distributions. Results: Compared to Normal Controls (NCs), patients with BD exhibited significantly higher entropy at the global level and within the Default Mode, Salience, and Somatosensory-Motor networks, indicating widespread network dedifferentiation (distributional flattening). These alterations were robust across different kernel widths and remained significant after rigorously controlling for head motion (Mean FD). Furthermore, manic symptom severity (YMRS) was negatively correlated with global entropy, suggesting a pathological “locking-in” or rigidity of specific neural circuits during manic states. Conclusions: The kernel-transformed FC entropy serves as a distribution-sensitive complement to conventional linear metrics. Our findings highlight network dedifferentiation as a key pathophysiological feature of BD and suggest this framework as a promising candidate metric for characterizing network dysregulation.
Zhang et al. (Tue,) studied this question.