High- and Low-Level Contextual Modeling for the Detection of Mild Traumatic Brain Injury

Abstract

Traumatic brain injury (TBI) can lead to long-term neurological decrements. While moderate and severe TBI are readily discernable from current medical imaging modalities, such as computed tomography and magnetic resonance imaging (MRI), mild TBI (mTBI) is difficult to diagnose from current routine imaging. At the present time there no routine computational methods for the evaluation of mild traumatic brain injury (mTBI) from magnetic resonance imaging (MRI). The development of automated analyses has been hindered by the subtle nature of mTBI abnormalities, which appear as low contrast MR regions. One solution to better identify mTBI injuries from MRI is to use high-level and low-level contextual information. We describe methods and results for using high-level contextual features using a Bayesian network that simulated the evolution of the mTBI injury over time. We also utilized low-level context to obtain more spatial (within the brain) information. The low-level context utilized a classifier to identify temporal information which was then integrated into the subsequent time point being evaluated. We found that both low-and high-level context provided novel information about the mTBI injury. These were in good agreement with manual methods. Future work could combine both low-and high-level context to provide more accurate mTBI segmentation. The results reported herein could ultimately lead to better identification of regions of mTBI injury and thus when treatments become available they can be directed for improved therapy.