Traumatic brain injury (TBI) is a leading cause of morbidity and mortality. Diffuse axonal injury (DAI), or lesions on the brain’s connective nerve fibers, is a common condition resulting from TBI. Previous research has found that the location and characteristics of diffuse axonal injury may contribute to different outcomes, including level of function and survival. In other fields of medicine, machine learning has revolutionized the ability of clinicians to predict patient outcomes by analyzing radiological imaging. For instance, machine learning has been used to successfully identify early signs of cancer.
Despite its prevalence, a limited number of TBI patients undergo diffuse axonal injury screening because it is difficult to diagnose and there is little clinical direction for those who are diagnosed. Magnetic resonance imaging (MRI) and computed tomography (CT) techniques have allowed clinicians to detect the incidence of diffuse axonal injury, yet no clear prognostic indicator has been established. To develop a definitive clinical indicator for diffuse axonal injury, a team of researchers used radiological images to create a model that predicts level of functioning and survival six months following injury.
The study reviewed MRI and CT scans of 38 adults with moderate to severe TBI that resulted in diffuse axonal injury. They also used a machine learning model to predict patient outcomes as either “favorable” or “unfavorable.” The model’s predictions were then compared to each patient’s clinical evaluation to assess the algorithm’s accuracy. The researchers found that their model had an overall accuracy of 57% in predicting the general favorability of patient outcomes. Their model showed high sensitivity specifically for favorable results—in other words, “favorable” outcomes determined by the algorithm were very consistent with the patient’s actual status. Conversely, the model’s detection of “unfavorable” results was only variably consistent with actual patient status.
Despite its limitations in precision, this study is the first reported use of machine learning for interpreting MRI to predict TBI outcomes for individuals with diffuse axonal injury. While more research needed to improve the efficiency of this technology, this advance in machine learning represents a critical first step in the expansion of predictive clinical measures for TBI.
Mohamed M, Alamri A, Mohamed M, et al. Prognosticating outcome using magnetic resonance imaging in patients with moderate to severe traumatic brain injury: a machine learning approach. Brain Injury. (January 2022).