A novel EEG-based major depressive disorder detection framework with two-stage feature selection

• DOI: 10.1186/s12911-022-01956-w
• Publication Date: 2022-08-06T18:02:49Z
• Authors: Yujie Li; Yingshan Shen; Xiaomao Fan; Xingxian Huang; Haibo Yu; Gansen Zhao; Wenjun Ma
• Keywords: Feature (linguistics)

Abstract Background Major depressive disorder (MDD) is a common mental illness, characterized by persistent depression, sadness, despair, etc., troubling people’s daily life and work seriously. Methods In this work, we present novel automatic MDD detection framework based on EEG signals. First of all, derive highly MDD-correlated features, calculating the ratio extracted features from signals at frequency bands between $$\beta$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>β</mml:mi> </mml:math> $$\alpha$$ <mml:mi>α</mml:mi> . Then, two-stage feature selection method named PAR presented with sequential combination Pearson correlation coefficient (PCC) recursive elimination (RFE), where advantages lie in minimizing searching space. Finally, employ widely used machine learning methods support vector (SVM), logistic regression (LR), linear (LNR) for merit interpretability. Results Experiment results show that our proposed achieves competitive results. The accuracy $$F_{1}$$ <mml:msub> <mml:mi>F</mml:mi> <mml:mn>1</mml:mn> </mml:msub> score are up to 0.9895 0.9846, respectively. Meanwhile, determination $$R^2$$ <mml:msup> <mml:mi>R</mml:mi> <mml:mn>2</mml:mn> </mml:msup> severity assessment 0.9479. Compared existing best 0.9840 $$F_1$$ 0.97, state-of-the-art performance. Conclusions Development can be potentially deployed into medical system aid physicians screen out patients.