EEG-based schizophrenia diagnosis using deep learning with multi-scale and adaptive feature selection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Schizophrenia, a severe mental illness, often presents with negative symptoms that are difficult to diagnose objectively.
• Current diagnostic methods for schizophrenia rely heavily on clinical experience, leading to potential misdiagnosis.

Purpose of the Study:

• To develop an objective and effective deep learning-based diagnostic method for schizophrenia using electroencephalogram (EEG) data.
• To address the challenge of detecting subtle brain wave patterns associated with schizophrenia.

Main Methods:

• Proposed a novel deep learning model: Cascaded Atrous Convolutional Network with Adaptive Weight Fusion (CA-AWFM).
• Utilized atrous convolutions for multi-scale temporal information extraction and a cascade network for progressive feature representation.
• Incorporated an Adaptive Weight Fusion Module (AWFM) for dynamic feature importance modification.

Main Results:

• The CA-AWFM model achieved a 99.5% accuracy rate in classifying schizophrenia from EEG data.
• Demonstrated superior performance compared to existing methods in schizophrenia detection.
• Effectively modeled local and global dependencies within EEG signals.

Conclusions:

• The CA-AWFM model shows high accuracy and potential for routine clinical use in early schizophrenia identification.
• The approach offers a promising objective tool for diagnosing schizophrenia, enabling timely intervention.
• Advanced deep learning techniques can significantly improve the accuracy of diagnosing complex neurological disorders like schizophrenia.