OBJECTIVE: Traumatic brain injury (TBI) often results in traumatic axonal injury (TAI). This can be difficult to identify using conventional imaging. Diffusion tensor imaging (DTI) offers a method of assessing axonal damage in vivo, but has previously mainly been used to investigate groups of patients. Machine learning techniques are increasingly used to improve diagnosis based on complex imaging measures. We investigated whether machine learning applied to DTI data can be used to diagnose white matter damage after TBI and to predict neuropsychological outcome in individual patients. METHODS: We trained pattern classifiers to predict the presence of white matter damage in 25 TBI patients with microbleed evidence of TAI compared to neurologically healthy age-matched controls. We then applied these classifiers to 35 additional patients with no conventional imaging evidence of TAI. Finally, we used regression analyses to predict indices of neuropsychological outcome for information processing speed, executive function, and associative memory in a group of 70 heterogeneous patients. RESULTS: The classifiers discriminated between patients with microbleeds and age-matched controls with a high degree of accuracy, and outperformed other methods. When the trained classifiers were applied to patients without microbleeds, patients having likely TAI showed evidence of greater cognitive impairment in information processing speed and executive function. The classifiers were also able to predict the extent of impairments in information processing speed and executive function. INTERPRETATION: The work provides a proof of principle that multivariate techniques can be used with DTI to provide diagnostic information about clinically significant TAI.
489 - 499
Adult, Anisotropy, Association Learning, Brain Injuries, Cognition Disorders, Executive Function, Female, Humans, Leukoencephalopathies, Logistic Models, Male, Memory, Middle Aged, Nerve Fibers, Myelinated, Neuropsychological Tests, Predictive Value of Tests, Young Adult