Predicting Outcome after Moderate-Severe TBI using a CT Head Deep Learning Model

Accurate outcome prediction following TBI can help with prognosis and determining health service needs. Computed tomography (CT) of the head is the standard for evaluating acute TBI, however, current CT classification methods correlate only with acute outcomes (e.g., need for brain surgery, death). This lack of an objective anatomical measure of injury severity is a gap in knowledge. Deep-learning (DL; also known as artificial intelligence) is capable of encoding imaging features invisible to the naked eye, or not apparent because of complex relationships among subtle features on digital images. In preliminary work, DL applied to acute digital CT scan data showed promise in providing prognostic information. For this study we will partner with our Digital Innovation Lab and other TBIMS centers to advance this research by combining clinical information from previously enrolled TBIMS National Database participants with their digital CT data to inform a DL model of anatomical injury severity. We hope this model will 1) more precisely correlate with clinical indicators of injury severity by providing a neuroanatomic biomarker of injury severity, 2) provide meaningful long-term prognostic information important to individuals with TBI and their families, and 3) guide clinical decision making and long-term rehabilitation care planning.