AI-powered Schizophrenia Screening with Consideration of Facial Feature and Privacy Protection

AI Summary

An AI screening framework integrates facial feature analysis with privacy protection to improve schizophrenia screening accuracy and efficiency.
Methodology: preprocess illumination; extract PCA features from whole, upper, and upper left face; classify using KNN, SVM, XGB with weighted ensemble.
Evaluation on patient and CAS PEAL-R1 datasets shows upper left face yields no statistically significant AUC loss versus whole face; robust ROC, PR, sensitivity, specificity.

TL;DR (AI Generated)

However, existing methods either prioritize privacy protection while neglecting facial features, which compromises screening efficiency, or focus on facial features without ensuring privacy protection, risking patient information leakage.

To address these limitations, this paper proposes an AI-powered schizophrenia screening framework that integrates facial feature analysis with privacy protection, aiming to enhance both screening accuracy and efficiency.

Specifically, we consider three facial regions, namely, the whole face, the upper face, and the upper left face, and apply illumination preprocessing to the collected images.

Basic summary

IEEE J Biomed Health Inform. 2026 Nov 13;PP. doi: 10.1109/JBHI.2026.3753632. Online ahead of print.

ABSTRACT

Schizophrenia is a complex mental disorder, and early identification and intervention are crucial for improving patients’ quality of life. Recently, the application of artificial intelligence (AI) technology in schizophrenia screening has attracted widespread attention. However, existing methods either prioritize privacy protection while neglecting facial features, which compromises screening efficiency, or focus on facial features without ensuring privacy protection, risking patient information leakage. To address these limitations, this paper proposes an AI-powered schizophrenia screening framework that integrates facial feature analysis with privacy protection, aiming to enhance both screening accuracy and efficiency. Specifically, we consider three facial regions, namely, the whole face, the upper face, and the upper left face, and apply illumination preprocessing to the collected images. Feature extraction is then performed using Principal Component Analysis (PCA). For classification, we employ three mainstream classifiers: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGB). Additionally, we propose a classifier weighted prediction model to further improve classification accuracy and robustness. Extensive experiments are conducted on two datasets, one with schizophrenia patients collected with smartphones at a hospital and the other with healthy individuals from the CAS PEAL-R1 dataset. The results demonstrate the effectiveness of our proposed scheme as measured by the Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves, confusion matrices, sensitivity, and specificity, with 95% confidence intervals, and show that using only the upper left face causes no statistically significant loss in AUC compared with the whole face. We also investigate the impact of classifier weights and the screening time. Deployment challenges and future research directions are discussed at the end of this paper.

PMID:43024545 | DOI:10.1109/JBHI.2026.3753632

Document this CPD