Quantum-inspired feature extraction model from EEG frequency waves for enhanced schizophrenia detection

Abstract

Purpose

Schizophrenia diagnosis remains challenging due to the reliance on subjective clinical assessments and the lack of robust, objective biomarkers. Current neuroimaging methods are often expensive, time-consuming, and may lack specificity, highlighting the need for the development of scalable and accurate diagnostic tools. This study investigates the feasibility of using electroencephalogram (EEG) frequency waves as biomarkers for the detection of schizophrenia, employing a quantum-based feature extraction methodology. The primary objective of this research is to develop an advanced detection methodology that integrates quantum-based feature extraction with sophisticated channel and feature selection techniques. This approach aims to enhance the accuracy and reliability of schizophrenia diagnosis by identifying the most informative EEG channels and features for classification purposes.

Methods

First, EEG frequency bands are extracted using the wavelet packet decomposition technique. Next, three channel selection algorithms prioritize channels based on the highest variance, power, and lowest coefficient of variation. The methodology involves applying discrete quantum analysis for feature extraction, followed by the extraction of statistical measures to create a comprehensive feature set. Feature selection is performed using Minimum Redundancy Maximum Relevance (mRMR) and ReliefF to retain the most relevant and non-redundant features. These features are then analyzed using various classification models, including AdaBoost, Support Vector Machine (SVM), Decision Tree (DT), and K-Nearest Neighbors (KNN).

Results

The findings of the study underscore the superior performance of the mRMR method when combined with variance and coefficient of variation-based channel selection techniques, particularly in the β and θ frequency bands. The KNN classifier achieves 100 % accuracy, sensitivity, and F1 score for the δ, θ, SMR, and β waves under optimal conditions. The mRMR method attains an average accuracy of 92.80 % for δ waves, 95.20 % for θ waves, 92.59 % for SMR waves, and 94.26 % for β waves when used in conjunction with coefficient of variation-based channel selection. In contrast, the ReliefF method demonstrates suboptimal performance in higher frequency bands, such as the γ wave, achieving an average accuracy of only 51.55 % when paired with variance-based channel selection.

Conclusion

The proposed methodology presents a promising approach to improving the accuracy and reliability of schizophrenia diagnosis.