[Diagnostic Performance of Automated Intracranial Hemorrhage Detection Software and Analysis of Volume and Hounsfield...

To compare the diagnostic performance of syngo.CT Brain Hemorrhage (BH) (Siemens Healthcare, Forchheim, Germany) with physician interpretations on noncontrast head CT in patients with suspected intracranial hemorrhage (ICH), and to elucidate the volume and Hounsfield unit (HU) metrics of false-positive (FP) cases.

We analyzed 509 noncontrast head CT examinations obtained for suspected ICH. Presence or absence of hemorrhage was determined by board-certified radiologists or clinicians. Among positive cases, hemorrhage subtypes were labeled as intraparenchymal hemorrhage (IPH), intraventricular hemorrhage (IVH), subdural hematoma (SDH), epidural hematoma (EDH), and subarachnoid hemorrhage (SAH). BH outputs were categorized as true positive (TP), false positive (FP), false negative (FN), or true negative (TN). For FP cases, lesion location and BH-derived quantitative indices-lesion volume and HU values (minimum, maximum, mean, and standard deviation)-were compared with TP cases.

BH achieved a sensitivity 100%, a specificity 82.2%, a FP rate 17.8%, and an FN rate 0%. TP cases comprised IPH 64, SAH 33, SDH 83, EDH 11, and IVH 14. FP findings were located in cerebral sulci (n=22), brain parenchyma (n=12), vessels (n=10), dura mater (n=5), and bone (n=5). All BH indices differed between FP and TP cases (p<0.01).

BH showed favorable diagnostic performance relative to physician interpretations, and FP cases exhibited statistically significant differences from TP cases in lesion volume and HU-based metrics.