Identifying patients with basal ganglia intracranial hemorrhage (ICH) at risk of hematoma expansion (HE) may better define selection criteria for early surgical evacuation. The aim of this study was to use automated radiographic feature extraction to improve risk stratification for basal ganglia ICH expansion. A single-center retrospective review was performed to identify patients with basal ganglia ICH between 2013 and 2024. ICH volumes were automatically segmented from the initial noncontrast computed tomography (CT) of the head using a custom-trained convolutional neural network. Features were quantified from the segmented ICH including stereotactic location, normalized volumetric CT density (nv-CTD, measured as mean ICH CT density divided by the background parenchymal CT density), volume, orientation, and border irregularity. HE was defined as an increase in hemorrhage volume of 10 mL or at a rate of 1.7 mL/h. A total of 108 patients (median age 55 years, 62% male) were included. HE occurred in 24 patients (22%) and was associated with shorter duration between symptom onset and initial CT (median 1 vs 3 hours, P = .006), a lower nv-CTD (median 2.0 vs 2.2, P = .011), and a positive spot sign (41% vs 5%, P < .001). nv-CTD was positively associated with time to presentation (R2 = 0.13, P < .001) and was negatively associated with HE in spot-sign-negative patients (median 2.0 vs 2.1, P = .016). Multivariate logistic regression modeling using nv-CTD and spot sign as inputs demonstrated improved diagnostic accuracy compared with that of the spot sign alone (area under the receiver operating characteristic curve 0.80 vs 0.68, P = .008). The area under the receiver operating characteristic curve of nv-CTD alone was 0.67 (95% CI: 0.56-0.78), which was statistically similar to that of the spot sign alone (0.68, 95% CI: 0.54-0.82) (P = .819). nv-CTD is a measure of bgICH acuity and can augment spot-sign bgICH expansion risk stratification.
Read full abstract