Detecting abnormal dynamic patterns of phase changes in schizophrenia from complex-valued fMRI data

Dynamic analysis has shown advantages in detecting psychiatric-related functional alterations using magnitude-only fMRI data. However, polarity dynamics in source phase derived from complex-valued fMRI data remain largely unexplored, though it offers additional information about brain networks.

We propose a framework for analyzing source phase polarity dynamics. We first extract dynamic source phase maps using independent component analysis from sliding-window complex-valued fMRI data. Then, we detect voxel-wise polarity changes between adjacent windows and compute their cumulative distributions. Next, we model the distributions with a piecewise linear function under constraints of a minimal segment length and polarity change stability. Finally, we form three dynamic patterns of phase polarity changes via clustering of the model parameters, and obtain pattern-specific sub-regions within activation maps.

We test the proposed framework using complex-valued fMRI data from schizophrenia patients and healthy controls. The dynamic patterns demonstrate that rapid polarity changes occur frequently in the central region of an activation map, occasionally appear in the intermediate region, but rarely occur at the edges. Cluster centroids and the number of voxels within each pattern-specific sub-region show significant group differences.

Our approach reveals a positively correlated relationship between the average number of polarity changes and activation magnitude, and provides unique sensitivity of the phase polarity to disease-related alterations beyond magnitude-based features.

This work provides new evidence of abnormal dynamic phase patterns in schizophrenia, offering insights into the underlying mechanisms of mental disorders.