Functional connectivity and graph theory of impaired central visual pathways in acute ischemic stroke based on functional magnetic resonance imaging.

Background: Stroke represents a major contributor to disability, resulting in functional impairments and imposing a societal burden. Resting-state functional connectivity (FC) indicates brain interactions, with dynamic alterations offering insights into cerebral function. This study used static and dynamic functional connectivity (sFC/dFC) and machine learning (ML) models to assess FC alterations in patients with acute ischemic stroke (AIS), aiming to identify connections that distinguish patients from healthy controls (HCs) and study their potential as biomarkers. Methods: A clinical trial took place at the Stroke Center of Shanghai Sixth People's Hospital Affiliated to Shanghai Jiao Tong University from March 2021 to September 2022 (trial registration date: 26 May 2021). A total of 22 patients were enrolled, 6 were lost to follow-up, and 26 HCs were recruited (10 males, 16 females, average age: 66.62±7.53 years). We performed a comparative analysis of whole-brain sFC among three groups: 12 patients and 26 HCs who completed the 3-month experiment, 12 patients and 26 HCs who participated in the 7-day experiment, and 12 patients who underwent both experiments. Whole-brain sFC analysis utilized the Yeo 7-network parcellation (Yeo-7) and 90 automated anatomical labeling (AAL) regions of interest (ROIs). The dFC analysis revealed two states: State 1 (weak, high-frequency), and State 2 (strong, low-frequency). Linear support vector machine (linear-SVM), radial basis function support vector machine (RBF-SVM), K-nearest neighbors (KNN), random forest (RF), and decision tree (TREE) models were trained using sFC features. Results: Patients with AIS had significantly altered ventral attention network (VAN) and default mode network (DMN) connectivity compared to HCs. Clinical assessments showed cognitive impairment at 7 days [Montreal Cognitive Assessment (MoCA): 19.42±10.64, P<0.001], improving at 3 months (MoCA: 20.58±4.89, P<0.05). Analysis of sFC revealed significant changes in different brain regions (P<0.05). dFC analysis identified two distinct states: a "high-frequency weakly connected" state at 7 days (63.61%) and a "low-frequency strong connection" state at 3 months (36.39%). ML models (RBF-SVM, RF, TREE) were utilized to identify optimal feature subsets, with RBF-SVM demonstrating superior performance [area under the curve (AUC): 0.85, accuracy: 85%]. Conclusions: Changes in FC among patients with AIS, particularly within the DMN, visual system (VIS), and limbic network (LIM) networks, may serve as possible biomarkers. ML models employing sFC characteristics are promising for stroke classification and prognosis prediction and improve the understanding of stroke-related neurological impairments.