A multimodal neuroimaging classifier for alcohol dependence.

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Machine Learning

Background:

• Multimodal neuroimaging offers potential for psychiatric disorder classification.
• Previous studies often failed to demonstrate the benefit of combining multiple neuroimaging modalities.
• Alcohol dependence has established neurobiological effects, making it suitable for multimodal analysis.

Purpose of the Study:

• To develop and evaluate a multimodal classification scheme for alcohol dependence.
• To assess the diagnostic accuracy of combining structural, functional task-based, and resting-state neuroimaging data.
• To determine if multimodal neuroimaging improves classification accuracy over single modalities.

Main Methods:

• A multimodal classification scheme was developed and applied to neuroimaging data.
• Data included structural, functional task-based, and resting-state MRI from alcohol-dependent patients (N=119) and controls (N=97).
• An optimized procedure for selecting modality-specific classifiers and an ensemble prediction method were employed.

Main Results:

• The multimodal classification scheme achieved 79.3% diagnostic accuracy for alcohol dependence.
• This accuracy surpassed the best individual modality, grey-matter density, by 2.7%.
• The performance gain depended on specific design choices, including classifier selection and ensemble weighting.

Conclusions:

• Combining multiple neuroimaging modalities can moderately enhance machine-learning-based diagnostic classification accuracy in alcohol dependence.
• Careful selection of classifiers and ensemble methods are critical for maximizing the benefits of multimodal data.
• This study demonstrates the value of integrated neuroimaging approaches for psychiatric diagnostics.