Cognitive Function Assessment Using a Virtual Reality Serious Game System in Patients With Stable Schizophrenia: Prospective Cohort Study.
Cognitive impairment is a core and enduring deficit in schizophrenia, severely affecting social functioning and quality of life. Traditional assessments such as the MATRICS Consensus Cognitive Battery face limitations in validity and engagement. Virtual reality (VR) serious games may offer an immersive alternative, and machine learning (ML) can uncover complex behavioral patterns. However, integrating VR-based assessment with ML for discriminating stable-phase schizophrenia remains unexplored.
This prospective cohort study aimed to examine whether a VR serious game ("Fruit Pioneer") can effectively assess cognitive function in stable schizophrenia, verify its correlation with the standard Brief Cognitive Assessment Tool for Schizophrenia (B-CATS), and test the discriminative capacity using ML models. We hypothesize that (1) patients with schizophrenia will show poorer VR game performance than healthy controls (HCs), (2) VR metrics will correlate with B-CATS scores, and (3) ML models will help classify patients with schizophrenia and HCs using VR data.
A total of 42 patients with stable schizophrenia and 65 HCs (aged 18-40 years) were enrolled. Exclusion criteria included color blindness, visual impairment, substance abuse, and comorbid chronic physical diseases. Finally, 39 patients with schizophrenia and 64 HCs were included. Materials included the VR serious game "Fruit Pioneer," B-CATS (Digital Symbol Substitution Test, Trail Making Test Part A, Trail Making Test Part B, and Animal Fluency), Simulator Sickness Questionnaire, and Game Experience Questionnaire. Data were collected via standardized VR gameplay and paper-based assessments. Logistic regression and a support vector machine (SVM) model were built using VR metrics.
Patients with schizophrenia performed worse on all B-CATS subtests (all P<.001). They also showed lower VR total scores (median 467, IQR 376-544 vs median 683, IQR 616-753; P<.001), longer reaction times (median 1.11, IQR 0.995-1.23 vs median 1.03, IQR 0.96-1.1; P=.006), lower gaze hit rates (median 0.515, IQR 0.442-0.554 vs median 0.552, IQR 0.497-0.592; P=.01), and higher bomb penalty scores (median 150, IQR 95-170 vs median 108, IQR 85-131; P=.002). In the schizophrenia group, VR metrics correlated with B-CATS results, whereas this relationship was minimal in HCs. Classification performance of the SVM (average area under the curve [AUC]=0.874, 95% CI 0.860-0.888) was comparable to logistic regression (average AUC=0.854, 95% CI 0.838-0.870).
This study demonstrates the innovative integration of a VR serious game with ML to assess cognitive function in stable schizophrenia. Unlike prior VR studies focused mainly on validation, our approach combines behavioral metrics with an SVM model, achieving effective classification. The findings support the potential of a scalable digital assessment correlated with standard tests. In clinical practice, this system may serve as an engaging alternative to traditional methods, facilitating long-term cognitive monitoring and personalized rehabilitation strategies.
This prospective cohort study aimed to examine whether a VR serious game ("Fruit Pioneer") can effectively assess cognitive function in stable schizophrenia, verify its correlation with the standard Brief Cognitive Assessment Tool for Schizophrenia (B-CATS), and test the discriminative capacity using ML models. We hypothesize that (1) patients with schizophrenia will show poorer VR game performance than healthy controls (HCs), (2) VR metrics will correlate with B-CATS scores, and (3) ML models will help classify patients with schizophrenia and HCs using VR data.
A total of 42 patients with stable schizophrenia and 65 HCs (aged 18-40 years) were enrolled. Exclusion criteria included color blindness, visual impairment, substance abuse, and comorbid chronic physical diseases. Finally, 39 patients with schizophrenia and 64 HCs were included. Materials included the VR serious game "Fruit Pioneer," B-CATS (Digital Symbol Substitution Test, Trail Making Test Part A, Trail Making Test Part B, and Animal Fluency), Simulator Sickness Questionnaire, and Game Experience Questionnaire. Data were collected via standardized VR gameplay and paper-based assessments. Logistic regression and a support vector machine (SVM) model were built using VR metrics.
Patients with schizophrenia performed worse on all B-CATS subtests (all P<.001). They also showed lower VR total scores (median 467, IQR 376-544 vs median 683, IQR 616-753; P<.001), longer reaction times (median 1.11, IQR 0.995-1.23 vs median 1.03, IQR 0.96-1.1; P=.006), lower gaze hit rates (median 0.515, IQR 0.442-0.554 vs median 0.552, IQR 0.497-0.592; P=.01), and higher bomb penalty scores (median 150, IQR 95-170 vs median 108, IQR 85-131; P=.002). In the schizophrenia group, VR metrics correlated with B-CATS results, whereas this relationship was minimal in HCs. Classification performance of the SVM (average area under the curve [AUC]=0.874, 95% CI 0.860-0.888) was comparable to logistic regression (average AUC=0.854, 95% CI 0.838-0.870).
This study demonstrates the innovative integration of a VR serious game with ML to assess cognitive function in stable schizophrenia. Unlike prior VR studies focused mainly on validation, our approach combines behavioral metrics with an SVM model, achieving effective classification. The findings support the potential of a scalable digital assessment correlated with standard tests. In clinical practice, this system may serve as an engaging alternative to traditional methods, facilitating long-term cognitive monitoring and personalized rehabilitation strategies.
Authors
Li Li, Zhuo Zhuo, Meng Meng, Zhao Zhao, Wu Wu, Yan Yan, Yue Yue, Sun Sun, Xiong Xiong, Cao Cao, Kou Kou, Yu Yu
View on Pubmed