Applying machine learning to dissociate between stroke patients and healthy controls using eye movement features obtained from a virtual reality task

Abstract

Conventional neuropsychological tests do not represent the complex and dynamic situations encountered in daily life. Immersive virtual reality simulations can be used to simulate dynamic and interactive situations in a controlled setting. Adding eye tracking to such simulations may provide highly detailed outcome measures, and has great potential for neuropsychological assessment. Here, participants (83 stroke patients and 103 healthy controls) we instructed to find either 3 or 7 items from a shopping list in a virtual super market environment while eye movements were being recorded. Using Logistic Regression and Support Vector Machine models, we aimed to predict the task of the participant and whether they belonged to the stroke or the control group. With a limited number of eye movement features, our models achieved an average Area Under the Curve (AUC) of .76 in predicting whether each participant was assigned a short or long shopping list (3 or 7 items). Identifying participant as either stroke patients and controls led to an AUC of .64. In both classification tasks, the frequency with which aisles were revisited was the most dissociating feature. As such, eye movement data obtained from a virtual reality simulation contain a rich set of signatures for detecting cognitive deficits, opening the door to potential clinical applications.