Research on drug addiction detection based on AR-TSNET with bimodal EEG-NIRS

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Traditional drug addiction assessment lacks objective physiological indicators and quantitative evaluation.
• Current methods rely on subjective psychological scales and self-reports.
• There is a need for more objective and reliable methods to assess addiction severity.

Purpose of the Study:

• To develop and validate a deep learning-based method for objective drug addiction assessment.
• To utilize multimodal neuroimaging data (EEG and NIRS) for enhanced accuracy.
• To evaluate the efficacy of a novel deep learning algorithm, AR-TSNET, in classifying addiction severity.

Main Methods:

• A visual trigger paradigm was used to elicit drug cravings in individuals with substance addiction.
• Electroencephalogram (EEG) and Near-Infrared Spectroscopy (NIRS) data were acquired from healthy individuals and those with drug addiction.
• A deep learning algorithm, AR-TSNET, employing feature-level fusion with Tception and Sception modules, attention mechanisms, and residual connections, was developed for classification.

Main Results:

• The proposed AR-TSNET model achieved a classification accuracy of 92.6% using k-fold cross-validation.
• Bimodal evaluation using both EEG and NIRS data demonstrated superior performance compared to single-modal approaches.
• Confusion matrix and ROC curve analyses confirmed the model's excellent performance in assessing drug addiction.

Conclusions:

• The developed deep learning approach using EEG and NIRS is a promising and effective method for objective drug addiction severity assessment.
• The multimodal neuroimaging approach offers higher information content and improved reliability over traditional methods.
• This objective, accurate, and reliable method has potential applications in addiction research and clinical practice, reducing reliance on subjective evaluations.