Machine learning algorithms revealed distorted retinal vascular branching in individuals with bipolar disorder

Abstract

BackgroundConverging evidence designate vascular vulnerability in bipolar disorder. The predisposition progresses into distortion in time, thus detection of the vascular susceptibility may help reducing morbidity and mortality. It was aimed to assess retinal fundus vasculature in cardiovascular risk-free patients with bipolar disorder. MethodsTotal of 68 individuals (38 patients with bipolar disorder, 30 healthy controls) were enrolled. In order to avoid from degenerative processes, participants were between 18 and 45 years of age, vascular risk factors were eliminated. Microscopic retinal fundus images were processed with machine learning algorithms (multilayer perceptron and support vector machine) and artificial neural network approaches. ResultsIn comparison to the healthy control group, the bipolar disorder group had lower number of breaking points (P < 0.001), lower number of curved vessel segments (P < 0.001). Total length of smooth vessels was longer (P = 0.040), and total length of curved vessel segments was significantly shorter (P < 0.001) than the control group. Vascular endothelial growth factor levels and gender were the confounders. There were significant correlations between vascular measures and serum lipid levels. LimitationsSample size was small and patients were on various medications. ConclusionsThese results indicate distortion in retinal vascular branching in bipolar disorder. Disrupted branching may reflect disturbed prosperity of retinal vascular plexus in patients with bipolar disorder. Alterations in the retinal vessels might be indicators of disruption in cerebral vascular system efficiency and thus neurovascular unit dysfunction in bipolar disorder.