Convolutional Neural Network Classification of Topographic Electroencephalographic Maps on Alcoholism

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging

Background:

• Alcohol use disorder (AUD) is a major global health concern, necessitating advanced diagnostic tools.
• Electroencephalography (EEG) offers a non-invasive method for brain activity assessment, but its application in AUD classification is underexplored.
• Limited research exists on using convolutional neural networks (CNNs) for classifying alcoholism based on topographic EEG signals.

Purpose of the Study:

• To develop and evaluate a CNN-based computational tool for classifying alcoholism using topographic EEG signals.
• To investigate the impact of dataset size on CNN classification accuracy for EEG data.
• To explore data augmentation techniques for improving EEG-based alcoholism detection.

Main Methods:

• An original dataset of EEG signals was collected from Brazilian participants during a language recognition task.
• Event-Related Potentials (ERPs) were transformed into topographic maps using statistical parameters over time.
• A CNN model was employed to classify these topographic EEG datasets.

Main Results:

• The study demonstrated the feasibility of using CNNs to classify topographic EEG patterns associated with alcohol abuse.
• Testing revealed that dataset size significantly affects CNN classification accuracy.
• A proposed data augmentation approach successfully increased the size of the topographic dataset, leading to improved classification accuracies.

Conclusions:

• CNNs show promise for classifying abnormal topographic EEG patterns indicative of alcohol abuse.
• The findings support the development of computational tools for objective alcoholism diagnosis using EEG.
• Data augmentation is a viable strategy to enhance the performance of CNNs in EEG-based alcoholism detection.