Objective: To investigate spontaneous neuronal activity changes in patients with central retinal artery occlusion (CRAO) using the resting-state functional magnetic resonance imaging (fMRI) and detect whether these brain functional alterations can represent an objective biomarker of clinical response using a machine learning algorithm.
Methods: Eighteen patients with CRAO and eighteen healthy controls (HCs) were recruited. The regional homogeneity (ReHo) method of resting-state fMRI was conducted to evaluate the synchronous brain activity alterations between two groups. Differences of ReHo values between two groups were compared using the independent two-sample t-test. The support vector machine algorithm was to distinguish patients of CRAO from HCs based on the two groups’ whole-brain ReHo patterns. The accuracy, sensitivity, and specificity for the classification were calculated. The classification performance was evaluated using the non-parametric permutation test.
Results: Compared to HCs, individuals with CRAO showed significantly lower ReHo in the right cerebellum and precuneus. Meanwhile, significant higher ReHo values were observed in the left superior temporal gyrus, postcentral gyrus, and precentral gyrus in the CRAO group compared to HCs. Furthermore, our results suggested that 77.78% individuals with CRAO could be successfully distinguished from HCs by the machine learning, with a sensitivity of 72.22% and a specificity of 83.33%, respectively. The area of receiver operating characteristic curve was calculated to be 0.85.
Conclusion: This study uncovered individuals with CRAO exhibited disturbed synchronous neuronal activities in multiple brain areas using neuroimaging techniques. The ReHo variability could distinguish individuals with CRAO from HCs with high accuracy.
Keywords: central retinal artery occlusion, functional magnetic resonance imaging, regional homogeneity, machine learning, support vector machine