Functional connectivity and graph theory of impaired central visual pathways in acute ischemic stroke based on fMRI

Abstract

Abstract In the study of this paper, we first performed the analysis of whole brain static functional connectivity, we divided the whole brain into 90 regions of interest (ROIs) by applying AAL mapping, we compared the whole brain static functional connectivity analysis of the 14 patients and 26 healthy volunteers (HC) who completed the 3-months experiment (3months), the 14 patients and 26 healthy volunteers who completed the 7-days experiment (7days), the 12 patients and the 12 patients who completed the 7-days experiment (7days), the 12 patients and the 12 patients who completed the 3-months experiment (7days), and the 12 patients and 26 healthy volunteers ( HC), 14 patients who completed the 7-day experiment (7days), and 14 patients who completed the 3-month experiment (3months) were analysed for whole-brain static functional connectivity in all three groups, and 90 ROIs were mapped to the Yeo7 functional network for analysis. sFC analyses revealed significant alterations in the patients' VAN, and DMN networks. Secondly, we performed dynamic functional connectivity analysis based on AAL mapping with the sliding window method separately, and identified two dynamic functional connectivity pattern characteristics, i.e., state 1 with a connectivity pattern dominated by high-frequency weak connectivity, and state 2 with a connectivity pattern dominated by low-frequency strong connectivity.Stroke patients spent significantly more time in state 1, and the number of state switches of the stroke patients in 7days significantly higher and were more likely to switch to the low-frequency strong connectivity mode state 2. Significant changes in connectivity were observed for DMN, VIS, FPN, and LIM. Finally, we built five machine learning models based on SFC features that differ between groups, namely linear support vector machine (SVM), radial basis function support vector machine (SVM-RBF), k nearest neighbours (KNN), random forest (RF), and decision tree (TREE). Based on the maximum AUC we identified the optimal feature subset and found that the SFC within the VIS, DMN, and LIM networks contributed significantly to the classification of AIS patients and HCs alike.The variation of FC within the VIS, DMN, and LIM networks may provide new insights into the neural mechanisms of AIS patients.