Review of feature selection techniques in Parkinson's disease using OCT-imaging data

Abstract

Several spectral-domain optical coherence tomography studies (OCT) reported a decrease on the macular region of the retina in Parkinson’s disease. Yet, the implication of retinal thinning with visual disability is still unclear.

Macular scans acquired from patients with Parkinson’s disease (n = 100) and a control group (n = 248) were used to train several supervised classification models. The goal was to determine the most relevant retinal layers and regions for diagnosis, for which univari- ate and multivariate filter and wrapper feature selection methods were used. In addition, we evaluated the classification ability of the patient group to assess the applicability of OCT measurements as a biomarker of the disease.