Deep learning-based multimodality classification of chronic mild traumatic brain injury using resting-state...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Mild traumatic brain injury (mTBI) presents a significant public health challenge.
• Objective diagnostic tools are needed for chronic mTBI assessment.
• Current diagnostic methods often lack sensitivity for long-term mTBI effects.

Purpose of the Study:

• To develop an automated classifier for distinguishing chronic mTBI patients from healthy controls.
• To explore the efficacy of deep learning frameworks using multimodal neuroimaging data.
• To identify discriminative brain features indicative of mTBI.

Main Methods:

• Utilized resting-state functional MRI (rs-fMRI) and positron emission tomography (PET) data.
• Developed a deep learning framework with autoencoders (AE) for feature extraction.
• Integrated single and multimodal imaging data for classification.

Main Results:

• Single neuroimaging modalities achieved classification accuracies of 79-91.67%.
• The multimodal model integrating rs-fMRI and PET data improved classification accuracy to 95.83%.
• Key discriminative features were identified in the default mode network, sensorimotor network, visual cortex, cerebellum, and limbic system.

Conclusions:

• Multimodal neuroimaging data significantly enhances the accuracy of mTBI classification.
• The deep learning approach shows promise for developing objective biomarkers for mTBI.
• This methodology could be valuable for clinical mTBI assessment and management.