A dual-task-embedded virtual reality system for intelligent quantitative assessment of cognitive processing speed

Abstract

Introduction: Processing speed is defined as the ability to quickly process information, which is generally considered as one of the affected cognitive functions of multiple sclerosis and schizophrenia. Paper-pencil type tests are traditionally used in the assessment of processing speed. However, these tests generally need to be conducted under the guidance of clinicians in a specific environment, which limits their application in cognitive assessment or training in daily life. Therefore, this paper proposed an intelligent evaluation method of processing speed to assist clinicians in diagnosis.

Methods: We created an immersive virtual street embedded with Stroop task (VR-Street). The behavior and performance information was obtained by performing the dual-task of street-crossing and Stroop, and a 50-participant dataset was established with the label of standard scale. Utilizing Pearson correlation coefficient to find the relationship between the dual-task features and the cognitive test results, and an intelligent evaluation model was developed using machine learning.

Results: Statistical analysis showed that all Stroop task features were correlated with cognitive test results, and some behavior features also showed correlation. The estimated results showed that the proposed method can estimate the processing speed score with an adequate accuracy (mean absolute error of 0.800, relative accuracy of 0.916 and correlation coefficient of 0.804). The combination of Stroop features and behavior features showed better performance than single task features.

Discussion: The results of this work indicates that the dual-task design in this study better mobilizes participants' attention and cognitive resources, and more fully reflects participants' cognitive processing speed. The proposed method provides a new opportunity for accurate quantitative evaluation of cognitive function through virtual reality.

Keywords: behavior data; cognitive processing speed; dual-task; evaluation; machine learning; virtual reality.

FIGURE 1

System framework.

FIGURE 2

Virtual scene overview. (A) Three examples of virtual environment. Left: VE of Dual-Task I with a color-matching task. Middle: VE of Dual-Task II with a word-matching task. Right: VE of Dual-Task III with an interference-matching task. (B) Schematic structure of the street scene. is the zebra crossing; is the start point; is the destination for crossing the street; indicates the width of the street; shows the area where participants are allowed to move. The red arrow in the figure is the orientation of the x-axis, while the blue arrow is the orientation of the y-axis. (C) System design.

FIGURE 3

Stroop task settings. (A) Color-matching task settings. (B) Word-matching task settings. (C) Interference-matching task settings.

FIGURE 4

The experimental diagram of the participants and the operation method of the handle.

FIGURE 5

Distribution of basic cognitive capacity test scores on the y-axis with respect to age on the x-axis. The horizontal line represents the mean score, and the vertical line represents the mean age. The subgraph on the right shows the distribution of subjects in different fractional segments.

FIGURE 6

Fitting result and Bland–Altman plot of the three regression models. (A) LASSO, (B) SVR, (C) XGBoost.

FIGURE 7

Performance of regression models using different feature combinations. (A) Mean MAE and the standard error of the mean (SEM) in parentheses. (B) Mean ACC and SEM in parentheses. *p 0.05. ^**p 0.01.

FIGURE 8

Performance of regression models using three dual-tasks separately. (A) Mean MAE and the standard error of the mean (SEM) in parentheses. (B) Mean ACC and SEM in parentheses.