Schizophrenia detection from electroencephalogram signals using image encoding and wrapper-based deep feature...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Schizophrenia is a severe mental illness affecting cognition and perception, impacting ~1% of the global population.
• Early diagnosis and treatment are crucial, yet the exact causes remain elusive, necessitating advanced diagnostic tools.
• Electroencephalogram (EEG) offers high temporal resolution for detecting subtle brain activity changes relevant to schizophrenia.

Purpose of the Study:

• To develop and validate a novel, highly accurate framework for detecting schizophrenia using EEG signals.
• To leverage deep learning and transfer learning for automated feature extraction and improved diagnostic performance.
• To enhance the efficiency and accuracy of schizophrenia detection compared to traditional methods.

Main Methods:

• A three-stage framework was proposed, starting with encoding EEG signals into scalogram images.
• Pre-trained deep learning models with transfer learning were employed for feature extraction from EEG images.
• An Average Subtraction wrapper-based feature selection method was introduced to reduce irrelevant features.

Main Results:

• The proposed framework achieved exceptional accuracy, reaching 99.67% on the M.S.U. dataset and 99.97% on the RepOD dataset.
• The method demonstrated superior performance over state-of-the-art results on both tested datasets.
• Automated feature selection significantly improved the speed and accuracy of schizophrenia detection.

Conclusions:

• The developed three-stage framework offers a highly accurate and efficient method for schizophrenia detection from EEG signals.
• Deep learning and transfer learning, combined with innovative feature selection, show significant promise for advancing neurological disorder diagnostics.
• This approach provides a valuable tool for improving the early detection and management of schizophrenia.