Related Experiment Video
Updated: May 25, 2026
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
A multimodal fNIRS-based machine learning model for symptom assessment and treatment response prediction in
Lei Cheng1, Xiaolu Xu1, Haosheng Yang1
1Department of Psychiatry, Henan Mental Hospital, the Second Affiliated Hospital of Henan Medical University, Xinxiang, China.
Objective:
Predicting early symptom severity and treatment response in schizophrenia is crucial for selecting optimal therapeutic strategies. This study aimed to develop machine learning (ML) models utilizing functional near-infrared spectroscopy (fNIRS) to predict clinical symptoms, cognitive function, and treatment responses.
Methods:
We enrolled 139 acutely ill schizophrenia patients and collected fNIRS data alongside clinical measures before and after a 4-week course of antipsychotic treatment. Based on the connectome-based predictive modeling (CPM) framework, regression models were constructed to predict symptom severity and cognitive function, while classification models were built to distinguish treatment response. All models were evaluated using a leave-one-out cross-validation.
Results:
The models demonstrated significant predictive power for clinical symptoms, cognitive function, and improvement (p < 0.001). However, the predictive efficacy for negative symptoms was relatively limited across all models (optimal model: r = 0.242, p < 0.05). All the classification models exhibited high sensitivity (>84%). The Support Vector Machine achieved an accuracy of 77.5%, and the Random Forest model achieved an area under the curve of 0.914.
Conclusion:
This study indicates that integrating fNIRS with ML not only deepens our understanding of the heterogeneity of schizophrenia but also enhances the accuracy of predicting disease severity and treatment outcomes. This provides a potential objective and reliable clinical tool for precision medicine in schizophrenia.
