Aberrant dynamic functional and effective connectivity changes of the primary visual cortex in patients with retinal detachment via machine learning

Abstract

Abstract Background: Retinal detachment (RD) is a prevalent and severe eye disease that often leads to vision loss. Previous research has indicated abnormal brain activity in individuals with RD. However, these studies solely focused on localized alterations in brain activity among individuals with RD, and it remains unclear if there are any changes in dynamic functional connectivity (dFC) and dynamic effective connectivity (dEC) in the primary visual cortex (V1) among individuals with RD. Aim: This study utilizes seed-based functional connectivity (FC) analysis and Granger causality analysis (GCA) to examine alterations in dynamic functional and effective connectivity in the V1 among patients with RD. Methods: The study involved 29 patients with RD and 30 healthy controls (HCs) who underwent resting-state functional magnetic resonance imaging (rs-fMRI) scans.Based on the seed regions in the V1, dynamic FC and GCA were conducted between the RD patients and HCs. To examine particular dFC and dEC states as well as associated temporal characteristics, the k-means clustering method was applied.The altered dFC and dEC values were selected as classification features and Support Vector Machine (SVM) classifiers were utilized to differentiate between patients with RD and HCs. Results: Compared to HCs, patients with RD displayed a significantly increased dFC between the right V1 and the temporal lobe, thalamus, frontal lobe, occipital lobe, angular gyrus, and cerebellum.Additionally, patients with RD exhibited significantly increased dFC between the left V1 and the parietal lobe.On the other hand, patients with RD showed a significantly decreased dFC between the left V1 and the cerebellum, amygdala, temporal lobe, and frontal lobe.Using the dynamic GCA algorithm, patients with RD showed a significant increase in dEC outflow from the right V1 to the frontal lobe, the caudate, the parietal lobule, and the angular gyrus.Patients with RD also exhibited a significant increase in dEC inflow to the right V1 from the temporal lobe, thalamus, the occipital lobe, and the parietal lobe.Additionally, patients with RD had significantly increased dEC outflow from the left V1 to the frontal lobe and the parietal lobe.Furthermore, patients with RD displayed a significant increase in dEC inflow to the left V1 from the occipital lobe.In contrast, patients with RD showed a significant decrease in dEC outflow from the left V1 to the occipital lobe. Lastly, patients with RD had significantly decreased dEC inflow to the left V1 from the occipital lobe and the postcentral gyrus[two-tailed, voxel-level p < 0.05, Gaussian random field (GRF) correction, cluster-level p < 0.05].After performing k-means clustering, it was observed that patients with RD predominantly displayed three dFC states and three or four dEC states.Depending on the region of interest (ROI), there are differences in the number of transitions(NT), frequency(F), and mean dwell time(MDT).The SVM model demonstrated accuracies of 0.712, 0.695, 0.525, 0.542, 0.593, and 0.458, along with corresponding areas under the curve (AUC) of 0.729, 0.786, 0.492, 0.561, 0.572, and 0, respectively, in distinguishing between individuals with RD and HCs based on the dFC/dEC values for the different ROI. Conclusion: Individuals with RD exhibited significant disruption in dFC/dEC between the V1 and multiple brain regions. The variability in dFC proved to distinguish individuals with RD from HCs with a high level of accuracy. These findings can contribute to the identification of potential neurological mechanisms underlying visual impairments in individuals with RD.