Automatic Diagnosis of Schizophrenia in EEG Signals Using CNN-LSTM Models
Afshin Shoeibi1, Delaram Sadeghi2, Parisa Moridian3
1Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran.
Frontiers in Neuroinformatics
|December 13, 2021
Summary
Deep learning methods, specifically CNN-LSTM, show high accuracy in diagnosing schizophrenia (SZ) using electroencephalography (EEG) signals. This approach surpasses conventional machine learning for automated SZ detection.
Area of Science:
- Neuroscience and Artificial Intelligence
- Computational Psychiatry
Background:
- Schizophrenia (SZ) is a complex mental disorder characterized by imbalances in brain chemistry, affecting thought, action, and emotional coordination.
- Accurate and early diagnosis of SZ is crucial for effective treatment and management.
Purpose of the Study:
- To develop and evaluate intelligent deep learning (DL) methods for automated schizophrenia diagnosis using electroencephalography (EEG) signals.
- To compare the performance of DL models against conventional machine learning techniques for SZ detection.
Main Methods:
- Utilized a dataset from the Institute of Psychiatry and Neurology in Warsaw, Poland.
- Preprocessed EEG signals by segmenting into 25s frames and normalizing using z-score or L2 norm.
- Implemented and compared conventional machine learning algorithms (SVM, k-NN, Decision Tree, Naïve Bayes, Random Forest, Extremely Randomized Trees, Bagging) and DL models (LSTMs, 1D-CNNs, 1D-CNN-LSTMs) with various activation functions.
Main Results:
- The 1D-CNN-LSTM architecture achieved the highest performance, utilizing the ReLU activation function with z-score and L2-combined normalization.
- The proposed CNN-LSTM model demonstrated a diagnostic accuracy of 99.25%, outperforming most previous studies.
- All simulations employed k-fold cross-validation with k=5 for robust evaluation.
Conclusions:
- Deep learning models, particularly the CNN-LSTM architecture, offer a highly accurate and effective approach for automated schizophrenia diagnosis via EEG.
- The developed DL method significantly advances the potential for objective and reliable SZ detection, surpassing traditional machine learning techniques.
