Machine learning-based classification of chronic traumatic brain injury using hybrid diffusion imaging.

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Machine Learning in Medicine

Background:

• Traumatic brain injury (TBI) can lead to progressive neuropathology and chronic impairments.
• There is a critical need for reliable biomarkers to detect and monitor TBI progression and severity.
• Current monitoring methods may not fully capture the extent of neurological damage.

Purpose of the Study:

• To evaluate data-driven analysis of diffusion tensor imaging (DTI) and neurite orientation dispersion imaging (NODDI) for developing TBI biomarkers.
• To compare the efficacy of DTI and NODDI biomarkers against conventional T1-weighted imaging.
• To determine if advanced imaging biomarkers can infer TBI symptom severity.

Main Methods:

• Developed machine learning models using hybrid diffusion imaging (HYDI) data from chronic TBI patients.
• Extracted relevant features from HYDI data.
• Trained and compared classification models based on DTI, NODDI, and T1-weighted imaging.

Main Results:

• Machine learning models based on DTI achieved 58.7-73.0% accuracy, and NODDI achieved 64.0-72.3% accuracy.
• These accuracies significantly outperformed conventional T1-weighted imaging, which had 51.7-56.8% accuracy.
• Feature selection and classification algorithms using diffusion-weighted imaging showed superior performance.

Conclusions:

• Machine learning-based analysis of DTI and NODDI significantly outperforms conventional T1-weighted imaging for TBI assessment.
• Advanced algorithms utilizing diffusion-weighted imaging features can effectively infer chronic brain injury symptoms.
• These findings support the development of novel biomarkers for improved TBI management.