Computer-Aided Assessment of Head Computed Tomography (CT) Studies in Patients with Suspected Traumatic Brain Injury

In this study, we sought to determine the accuracy of a computer algorithm that automatically assesses head computed tomography (CT) studies in patients with suspected traumatic brain injury (TBI) for features of intracranial hemorrhage and mass effect, employing a neuroradiologist's interpretation as the gold standard. To this end, we designed a suite of computer algorithms that evaluates in a fully automated fashion the presence of intracranial blood and/or mass effect based on the following CT findings: (1) presence or absence of a subdural or epidural hematoma, (2) presence or absence of subarachnoid hemorrhage, (3) presence or absence of an intraparenchymal hematoma, (4) presence or absence of clinically significant midline shift (>= 5 mm), and (5) normal, partly effaced, or completely effaced basal cisterns. The algorithm displays abnormal findings as color overlays on the original head CT images, and calculates the volume of each type of blood collection, the midline shift, and the volume of the basal cisterns, based on the above-described features. Thresholds and parameters yielding optimal accuracy of the computer algorithm were determined using a development sample of 33 selected, nonconsecutive patients. The software was then applied to a validation sample of 250 consecutive patients evaluated for suspicion of acute TBI at our institution in 2006 -2007. Software detection of the presence of at least one noncontrast CT (NCT) feature of acute TBI demonstrated high sensitivity of 98% and high negative predictive value (NPV) of 99%. There was actually only one false negative case, where a very subtle subdural hematoma, extending exclusively along the falx, was diagnosed by the neuroradiologist, while the case was considered as normal by the computer algorithm. The software was excellent at detecting the presence of mass effect and intracranial hemorrhage, but showed some disagreements with the neuroradiologist in quantifying the degree of mass effect and characterizing the type of intracranial hemorrhage. In summary, we have developed a fully automated computer algorithm that demonstrated excellent sensitivity for acute intracranial hemorrhage and clinically significant midline shift, while maintaining intermediate specificity. Further studies are required to evaluate the potential favorable impact of this software on facilitating workflow and improving diagnostic accuracy when used as a screening aid by physicians with different levels of experience.