Temporal variability analysis of frequency-optimized Parkinsonian mild cognitive impairment’s microstate network

The quantification and application of dynamic changes in brain functional connectivity networks can contribute to a better understanding of brain diseases such as Parkinson’s disease (PD) and to provide better prognostic indicators or auxiliary diagnostic tools clinically. In this study, we first constructed a convolutional neural network and used a gradient-weighted class activation mapping method to determine the characteristic frequency bands of early Parkinson’s mild cognitive impairment (ePD-MCI), then recognized the dynamic functional connectivity networks based on frequency-optimized electroencephalography (EEG) microstates and calculated its temporal variability. The experimental results indicate that the dynamic temporal variability of microstate-based brain networks can reflect pathological changes in the ePD-MCI brain. Furthermore, we also observed that the temporal variability of the ePD-MCI brain networks has a specific distribution pattern in the brain regions, which can quantify the degree of mild cognitive impairment. The temporal variability of a certain microstate class in the right frontal lobe, right central region, left occipital lobe, left parietal lobe, and left posterior temporal lobe can identify ePD-MCI. Our work provides innovative methodological support for future brain network studies and provides deeper insights into the spatiotemporal interaction patterns of brain activity and its changes in early PD.