Abstract Background and aims The interpretation of the five common intracranial hemorrhage (ICH) subtypes can be augmented by image-based segmentation information. The objective of this study is to test whether there is a relationship between bleed volume and the lateral ventricle volume (LVV) across each of the ICH subtypes. Methods RSNA ICH Detection Challenge 2019 CT imaging data (n=22061) were reconstructed. ICH subtypes were provided, namely: epidural (EDH), acute subdural (ASDH), subarachnoid (SAH), intraparenchymal (IPH) and intraventricular (IVH) hemorrhage. Cases were coded as single or mixed subtypes. We utilized an in-house dual-segmentation AI tool (VB-AI) to estimate both the ICH bleed and log LV volumes (LVV) (Figure 1). Data were split into 10 subgroups or folds. Linear relationship and slope values were estimated between ICH and LVV across 10-folds of the data. Results All bleed subtypes showed a small negative association between ICH and LVV, with the exception of IVH that showed a moderate negative relationship (Table 1). The findings were consistent for single- and mixed subtypes. Figure 2 shows the linear regression slope estimates across the 10 folds. One-way ANOVA resulted in a significant difference between subtypes (F-statistic=9.94, p-value 0.001). Conclusions Across the ICH subtypes, we found mild and moderate associations between ICH volume and LVV. Patients with IVH had the strongest negative association, which was expected due to the direct effect of the bleed on the fluid-filled ventricles. These findings speak to the face validity of the VB-AI tool for ICH research. Conflict of interest All authors have nothing to disclose Figure 1 - belongs to Methods Table 1 - belongs to Results Figure 2 - belongs to Results
Vries et al. (Fri,) studied this question.