Covariance-Eigenvalue Analysis of Multichannel EEG for Structural Characterization and Subject-Independent Schizophrenia Detection

Abstract This study investigates covariance-matrix eigenvalue representations as compact features for subject-independent schizophrenia classification using multichannel resting-state EEG.The analysis considers the covariance eigenvalue spectrum as a measure of the second-order arrangement of multichannel variance and evaluates it from many complimentary perspectives, including frequency, temporal, duration and structural elements. Five covariance-eigenvalue representations were created for frequency ranges of Delta (0.5-4 Hz), Theta (4-8 Hz), Alpha (8-13 Hz), Beta (13-30 Hz), and Gamma (30-45 Hz), resulting in 80 eigenvalue characteristics per participant. Comprehensive, standardized, condensed and sub band representations were tested using thorough subject-level leave-one-subject-out (LOSO) cross-validation. The analysis also considered temporal instability, band-wise eigenvalue differences, and singular-value-decomposition-based measures of structural dimensionality.The last locked 10-s five-band covariance-eigenvalue model with XGBoost achieved an accuracy of 89.29%, a balanced accuracy of 88.97%, an ROC-AUC of 90.71%, a sensitivity of 93.33%, and a specificity of 84.62%. Statistical verification was conducted with subject-specific bootstrap confidence intervals and paired exact McNemar tests with Holm correction. After Holm correction, XGBoost showed substantial differences in paired prediction errors with SVM-RBF and kNN, but no statistically significant differences with Random Forest and Logistic Regression.Model interpretation was performed using SHAP (SHapley Additive exPlanations). To assess whether dimensionality reduction could improve performance over the full 80-feature representation, nested SHAP/ANOVA-based feature-selection experiments and a localized sensitivity analysis for (K=55- 65) features were conducted. Feature selection did not improve the primary LOSO accuracy (88.10% vs. 89.29%) or balanced accuracy (87.86% vs. 88.97%), although a modest increase in ROC-AUC was observed (0.9151 vs. 0.9071).The results of the study support the covariance-eigenvalue spectrum as a compact statistical model for subject-independent EEG assessment. However, independent-cohort validation is necessary before any clinical or biomarker claims can be made.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-12
DOI
https://doi.org/10.21203/rs.3.rs-11483338/v1
Primary Topic
EEG and Brain-Computer Interfaces
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Covariance-Eigenvalue Analysis of Multichannel EEG for Structural Characterization and Subject-Independent Schizophrenia Detection

Ratheesh Sudheerbabu, Abhishek S, Vinayakumar B
Research Square
EEG and Brain-Computer Interfaces
preprint

Covariance-Eigenvalue Analysis of Multichannel EEG for Structural Characterization and Subject-Independent Schizophrenia Detection

Ratheesh Sudheerbabu, Abhishek S, Vinayakumar B
preprint en

Abstract

Abstract This study investigates covariance-matrix eigenvalue representations as compact features for subject-independent schizophrenia classification using multichannel resting-state EEG.The analysis considers the covariance eigenvalue spectrum as a measure of the second-order arrangement of multichannel variance and evaluates it from many complimentary perspectives, including frequency, temporal, duration and structural elements. Five covariance-eigenvalue representations were created for frequency ranges of Delta (0.5-4 Hz), Theta (4-8 Hz), Alpha (8-13 Hz), Beta (13-30 Hz), and Gamma (30-45 Hz), resulting in 80 eigenvalue characteristics per participant. Comprehensive, standardized, condensed and sub band representations were tested using thorough subject-level leave-one-subject-out (LOSO) cross-validation. The analysis also considered temporal instability, band-wise eigenvalue differences, and singular-value-decomposition-based measures of structural dimensionality.The last locked 10-s five-band covariance-eigenvalue model with XGBoost achieved an accuracy of 89.29%, a balanced accuracy of 88.97%, an ROC-AUC of 90.71%, a sensitivity of 93.33%, and a specificity of 84.62%. Statistical verification was conducted with subject-specific bootstrap confidence intervals and paired exact McNemar tests with Holm correction. After Holm correction, XGBoost showed substantial differences in paired prediction errors with SVM-RBF and kNN, but no statistically significant differences with Random Forest and Logistic Regression.Model interpretation was performed using SHAP (SHapley Additive exPlanations). To assess whether dimensionality reduction could improve performance over the full 80-feature representation, nested SHAP/ANOVA-based feature-selection experiments and a localized sensitivity analysis for (K=55- 65) features were conducted. Feature selection did not improve the primary LOSO accuracy (88.10% vs. 89.29%) or balanced accuracy (87.86% vs. 88.97%), although a modest increase in ROC-AUC was observed (0.9151 vs. 0.9071).The results of the study support the covariance-eigenvalue spectrum as a compact statistical model for subject-independent EEG assessment. However, independent-cohort validation is necessary before any clinical or biomarker claims can be made.

Research Square
Middle East College (OM), Amrita Vishwa Vidyapeetham (IN)
Amrita Vishwa Vidyapeetham University
Good health and well-being
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon