Diagnosis of Schizophrenia Using Feature Extraction from EEG Signals Based on Markov Transition Fields and Deep...

Area of Science:

• Neuroscience and Artificial Intelligence
• Biomedical Signal Processing

Background:

• Diagnosing schizophrenia via Electroencephalograph (EEG) signals is difficult due to subtle signal differences.
• Deep learning models, particularly those successful in image recognition, offer potential for improved diagnostic accuracy.
• Transforming 1D EEG signals into 2D representations facilitates image-based analysis.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for schizophrenia diagnosis using EEG data.
• To investigate the efficacy of Markov Transition Fields (MTF) in converting EEG signals into 2D images.
• To compare the performance of a deep learning pipeline against a traditional machine learning pipeline for classification.

Main Methods:

• EEG signals were converted into 2D images using the Markov Transition Field (MTF) to capture temporal and statistical dynamics.
• A pre-trained VGG-16 model was utilized for feature extraction from the generated 2D EEG images.
• Two classification pipelines were implemented: 1) Support Vector Machine (SVM) and 2) a deep learning approach with an autoencoder and neural network.

Main Results:

• The deep learning pipeline achieved a highest classification accuracy of 98.51% and 100% recall.
• The Support Vector Machine (SVM) pipeline demonstrated strong performance with a best accuracy of 96.28% and 97.89% recall.
• The study utilized the open-access Schizophrenia EEG database from MV Lomonosov Moscow State University.

Conclusions:

• The proposed deep learning method, employing MTF for EEG signal transformation, is highly effective for schizophrenia diagnosis.
• This approach offers a biomimetic strategy for pattern recognition and decision-making in neurological disorder diagnostics.
• The high accuracy and recall suggest significant potential for clinical application in early schizophrenia detection.