Related Experiment Video
Updated: Aug 5, 2026
05:19
Closed-Loop Neurostimulation for Biomarker-Driven, Personalized Treatment of Major Depressive Disorder
Published on: July 7, 2023
A Decision Support System for Major Depressive Disorder Detection Using EMD and CNN-BiLSTM-Based Analysis of Human
Nadide Gulsah Gulenc1, Mahmut Ozturk2
1Department of Biomedical Engineering, Çorlu Engineering Faculty, Tekirdag Namık Kemal University, Tekirdag, Turkiye; Department of Biomedical Engineering, Institute of Graduate Studies, Istanbul University - Cerrahpasa, Istanbul, Turkiye.
Summary
This study introduces a novel deep learning approach using Empirical Mode Decomposition (EMD) and a CNN-BiLSTM model for accurate, noninvasive diagnosis of Major Depressive Disorder (MDD) from speech signals.
Area of Science:
- Computational linguistics
- Artificial intelligence in healthcare
- Speech signal processing
Background:
- Major Depressive Disorder (MDD) significantly impacts speech characteristics.
- Current diagnostic methods for MDD lack noninvasive, objective, and automated approaches.
- Speech analysis offers a promising avenue for early MDD detection.
Purpose of the Study:
- To develop and validate a novel hybrid deep learning model for the early diagnosis of MDD using speech signals.
- To investigate the efficacy of Empirical Mode Decomposition (EMD) in enhancing speech feature extraction for MDD classification.
- To compare the proposed model's performance against established machine learning techniques.
Main Methods:
- Speech signals were decomposed into intrinsic mode functions (IMFs) using EMD.
- A comprehensive set of 325 acoustic features were extracted from raw and IMF-decomposed speech segments.
- A hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-BiLSTM) model was employed for feature classification.
Main Results:
- The EMD-based adaptive features significantly improved classification performance compared to features extracted without EMD.
- The proposed CNN-BiLSTM model achieved 94.7% accuracy and an average F1-score of 0.95.
- The method demonstrated superior performance over existing MFCC or spectrogram-based studies.
Conclusions:
- The developed EMD-based CNN-BiLSTM model provides a highly accurate, stable, interpretable, and generalizable method for MDD diagnosis from speech.
- This approach offers a noninvasive, objective, and automated tool for early MDD detection.
- A clinical decision support interface was developed to facilitate the practical application of this technology.
