Specific static and dynamic functional network connectivity changes in thyroid-associated ophthalmopathy and it predictive values using machine learning.

Background: Thyroid-associated ophthalmopathy (TAO) is a prevalent autoimmune disease characterized by ocular symptoms like eyelid retraction and exophthalmos. Prior neuroimaging studies have revealed structural and functional brain abnormalities in TAO patients, along with central nervous system symptoms such as cognitive deficits. Nonetheless, the changes in the static and dynamic functional network connectivity of the brain in TAO patients are currently unknown. This study delved into the modifications in static functional network connectivity (sFNC) and dynamic functional network connectivity (dFNC) among thyroid-associated ophthalmopathy patients using independent component analysis (ICA). Methods: Thirty-two patients diagnosed with thyroid-associated ophthalmopathy and 30 healthy controls (HCs) underwent resting-state functional magnetic resonance imaging (rs-fMRI) scanning. ICA method was utilized to extract the sFNC and dFNC changes of both groups. Results: In comparison to the HC group, the TAO group exhibited significantly increased intra-network functional connectivity (FC) in the right inferior temporal gyrus of the executive control network (ECN) and the visual network (VN), along with significantly decreased intra-network FC in the dorsal attentional network (DAN), the default mode network (DMN), and the left middle cingulum of the ECN. On the other hand, FNC analysis revealed substantially reduced connectivity intra- VN and inter- cerebellum network (CN) and high-level cognitive networks (DAN, DMN, and ECN) in the TAO group compared to the HC group. Regarding dFNC, TAO patients displayed abnormal connectivity across all five states, characterized by notably reduced intra-VN connectivity and CN connectivity with high-level cognitive networks (DAN, DMN, and ECN), alongside compensatory increased connectivity between DMN and low-level perceptual networks (VN and basal ganglia network). No significant differences were observed between the two groups for the three dynamic temporal metrics. Furthermore, excluding the classification outcomes of FC within VN (with an accuracy of 51.61% and area under the curve of 0.35208), the FC-based support vector machine (SVM) model demonstrated improved performance in distinguishing between TAO and HC, achieving accuracies ranging from 69.35 to 77.42% and areas under the curve from 0.68229 to 0.81667. The FNC-based SVM classification yielded an accuracy of 61.29% and an area under the curve of 0.57292. Conclusion: In summary, our study revealed that significant alterations in the visual network and high-level cognitive networks. These discoveries contribute to our understanding of the neural mechanisms in individuals with TAO, offering a valuable target for exploring future central nervous system changes in thyroid-associated eye diseases.