MuST-GCN: Multiscale and Hybrid Spatial Temporal Graph Convolutional Network for Accurate Identification of Alcohol...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alcohol Use Disorder (AUD), including Alcohol Abuse (AA) and Alcohol Dependence (AD), is a complex brain condition with overlapping symptoms, complicating diagnosis and treatment.
• Psychological differences, such as poor behavioral control in AA and affective lability in AD, manifest in distinct brain activity patterns.

Purpose of the Study:

• To develop an advanced computational model for accurately differentiating between AA and AD within AUD using electroencephalogram (EEG) signals.
• To leverage unique psychological and neurophysiological markers for improved diagnostic accuracy in AUD subtypes.

Main Methods:

• Introduction of the Multiscale and Hybrid Spatial Temporal Graph Convolutional Network (MuST-GCN) model for EEG feature extraction.
• Utilizing a Multiscale Feature Extraction (MSFE) module with Graph Convolutional Networks (GCN) to analyze brain region connectivity.
• Employing a Hybrid Spatial Temporal Memory (HSTM) module with attention mechanisms to refine features and enhance classification.

Main Results:

• MuST-GCN achieved high classification accuracies of 90.74% and 99.99% on two datasets via five-fold cross-validation.
• The model demonstrated superior performance in identifying AA, AD, and AUD compared to existing diagnostic methods.
• Refined feature representation by HSTM module effectively reduced overfitting and improved multiclass classification.

Conclusions:

• MuST-GCN offers a robust and accurate method for distinguishing between Alcohol Abuse and Alcohol Dependence using EEG data.
• The model's ability to identify AUD subtypes holds significant potential for guiding personalized treatment strategies.
• Publicly accessible code facilitates further research and clinical application of this advanced diagnostic tool.