Combining Results of Different Oculometric Tests Improved Prediction of Parkinson's Disease Development

摘要

In this text we compare the measurement results of reflexive saccades and antisaccades of patients with Parkinson's Disease (PD), trying to determine the best settings to predict the Unified Parkinson's Disease Rating Scale (UPDRS) results. After Alzheimer's disease, PD statistically is the second one and until today, no effective therapy has been found. Luckily, PD develops very slowly and early detection can be very important in slowing its progression. In this experiment we examined the reflective saccades (RS) and antisaccades (AS) of 11 PD patients who performed eye-tracking tests in controlled conditions. We correlated neurological measurements of patient's abilities described by the Unified Parkinson's Disease Rating Scale (UPDRS) scale with parameters of RS and AS. We used tools implemented in the Scikit-Learn for data preprocessing and predictions of the UPDRS scoring groups [1]. By experimenting with different datasets we achieved best results by combining means of RS and AS parameters into computed attributes. We also showed, that the accuracy of the prediction increases with the number of such derived attributes. We achieved 89% accuracy of predictions and showed that computed attributes have 50% higher results in the feature importance scoring than source parameters. The eye-tracking tests described in this text are relatively easy to carry out and could support the PD diagnosis.