Comparison of agreement between semi-automated software and the ABC/2 method in quantifying hematoma and perihematomal edema volumes in intracerebral hemorrhage

Reliable quantification of intracerebral hemorrhage (ICH) and perihematomal edema (PHE) is essential for clinical research, yet inter-method agreement remains unclear. We compared agreement for ICH and PHE volumes among three semi-automated software platforms as well as between ABC/2 and the software-derived average, assessing the effects of hematoma morphology and ICH volume. We analyzed 153 ICH patients who underwent non-contrast CT within 24 hours. ICH and PHE volumes were measured using ABC/2 and three platforms (3D Slicer, Analyze, ITK-SNAP). The software-derived average served as a composite comparator for ABC/2. Agreement was assessed using intraclass correlation coefficients (ICCs) and Bland-Altman analysis. Proportional bias was assessed by regression. Subgroups included hematoma morphology and ICH volume. Across pairwise comparisons, the three platforms showed excellent ICH agreement (ICCs >0.99; 95% limits of agreement (LOA): −3.65 to 2.39 mL). For PHE, despite excellent ICCs (>0.90), the 95% LOA were wider (−5.98 to 6.39 mL). Compared with the software-derived average, ABC/2 yielded higher ICH (+4.15 mL) and PHE (+0.96 mL) volumes, with significant proportional bias (p < 0.001). For ICH, ABC/2 bias was greater in irregular (7.50 vs 2.06 mL) and larger (10.25 vs 1.92 mL) hematomas, whereas PHE bias did not differ significantly across subgroups. The three software platforms showed excellent agreement for ICH volume, whereas PHE measurements should be interpreted cautiously because of the wide LOA. ABC/2 yielded higher volumes than the software-derived average, particularly in irregular and larger hematomas, which may affect classification near trial eligibility thresholds.