Related Experiment Video
Updated: Jun 5, 2026
08:51
Statistical Modelling of Cortical Connectivity Using Non-invasive Electroencephalograms
Published on: November 1, 2019
DCM-ML: An Electroencephalography-based Classifier for Early Diagnosis of Schizophrenia Based on Dynamic Connectivity
Seyed Abolfazl Valizadeh1, Marcus Cheetham2, Alireza Mohammadi3
1Student Research Committee, Baqiyatallah University of Medical Sciences, Tehran, Iran.
Basic and Clinical Neuroscience
|June 4, 2026
Summary
This study introduces a machine learning model using event-related potentials (ERPs) to accurately diagnose schizophrenia (SZ). The framework shows high accuracy and robustness, offering a promising non-invasive tool for early detection.
Area of Science:
- Neuroscience
- Computational Psychiatry
- Biomedical Engineering
Background:
- Early schizophrenia (SZ) diagnosis is challenging due to subjective assessments and symptom variability.
- Objective, scalable, and non-invasive diagnostic tools are needed to complement traditional methods.
- Dynamic connectivity matrices (DCMs) from event-related potentials (ERPs) offer a potential objective measure.
Purpose of the Study:
- To develop and validate a machine learning (ML) framework for schizophrenia classification using ERP-derived DCMs.
- To assess the accuracy, robustness, and generalizability of the proposed ML model.
- To identify key neurophysiological markers indicative of schizophrenia.
Main Methods:
- Utilized ERP data from 49 schizophrenia patients and 32 healthy controls from an anonymized public dataset.
- Computed 64x64 directional connectivity matrices using Granger causality to analyze inter-electrode information flow.
- Trained a random forest (RF) classifier on 2,777 significant connectivity differences, employing balanced subsets and evaluating robustness against white Gaussian noise.
Main Results:
- The RF classifier achieved high diagnostic accuracy (99.24%), sensitivity (98.34%), specificity (99.73%), and F1-score (98.91%).
- The model demonstrated robustness against noise (92% F1-score at 45% noise) and overfitting.
- Key discriminators included central, occipito-parietal, and inferior regions, suggesting disrupted fronto-temporal and sensory integration networks.
Conclusions:
- ML-driven ERP connectivity analysis is a feasible, non-invasive tool for early schizophrenia detection.
- The model exhibits strong generalizability, interpretability, and clinical scalability, outperforming deep learning approaches.
- Fronto-central and occipito-parietal connectivity patterns are diagnostically relevant; further validation on diverse cohorts is warranted.