Machine learning classification of chronic traumatic brain injury using diffusion tensor imaging and NODDI: A...

Area of Science:

• Neuroimaging
• Diffusion MRI
• Traumatic Brain Injury

Background:

• Traumatic brain injury (TBI) is linked to white matter (WM) abnormalities.
• Fractional anisotropy (FA) has previously classified acute TBI with 75.50% AUC.
• The efficacy of FA for chronic TBI classification remains unclear.

Purpose of the Study:

• To replicate and extend previous findings on FA for classifying chronic TBI.
• To investigate the utility of neurite orientation dispersion and density imaging (NODDI) metrics.
• To assess classification accuracy using FA and NODDI metrics in incarcerated men with and without chronic TBI history.

Main Methods:

• Employed a linear support vector machine (SVM) classifier.
• Utilized fractional anisotropy (FA) and NODDI metrics (orientation dispersion (ODI), isotropic volume (Viso)) as features.
• Classified 160 incarcerated men (80 with, 80 without chronic TBI history).

Main Results:

• Overall classification rates were high when incorporating FA and NODDI ODI metrics (AUC: 82.50%).
• NODDI-based metrics demonstrated superior performance: ODI achieved 85.00% sensitivity, and Viso achieved 82.50% specificity.
• The study successfully replicated and extended previous findings.

Conclusions:

• Multiple diffusion MRI metrics reliably differentiate individuals with and without self-reported chronic TBI history.
• NODDI metrics offer enhanced sensitivity and specificity for chronic TBI classification.
• Diffusion MRI holds promise for identifying individuals with a history of TBI.