A virtual reality simulation: classifying stroke patients from healthy controls based on wayfinding features

Abstract

Common cognitive assessment methods often lack ecological validity for assessing complex daily life situations. Virtual reality provides the opportunity to stablish controlled and complex daily life-like environments, while capturing the real-time data of patients. In this study 66 stoke patients and 102 healthy ontrols were subjected to a shopping task inside a Virtual Supermarket (VS) simulation. Movement properties were derived from the coordinate coordinate data that resulted from the VS simulation, and 199 novel way-finding features were extracted. For two separate task types of the VS simulation: a short and a long shopping list, a Logistic Regression model was trained on the optimal subset of wayfinding features. With these models, stroke patients could be distinguished from control subjects, with a performance of an Area Under the Receiver Operating Characteristics (AUROC) of 0.89 with respect to the short list task, and an AUROC of 0.80 for the long list task. The results of this research, suggest a great potential for combining VR simulations with machine learning techniques for innovative cognitive assessment methods, where patients can be exposed to more complex and ecological valid environments