Background Automated CT perfusion (CTP) software is widely used to select patients with large-vessel occlusion stroke for endovascular thrombectomy (EVT), yet treatment thresholds are often applied as if outputs from different platforms are interchangeable. This study compared RAPID and Brainomix CTP outputs in the same thrombectomy cohort.
Methods We performed a retrospective observational study of patients with large vessel occlusion who underwent CTP imaging. Paired volumetric data were analyzed using both RAPID and Brainomix algorithms. Inter-platform agreement was assessed using intraclass correlation coefficients (ICC), Pearson correlation (r), and Bland-Altman analysis. A simplified threshold-based reclassification was performed using the volumetric core cutoffs from DEFUSE-3 (<70 mL) and DAWN (<50 mL); this assessed only the volumetric criterion and does not represent full reconstruction of trial eligibility.
Results Our study found 328 EVT patients had complete paired outputs. Median core volumes were 4 mL (RAPID) vs 10 mL (Brainomix); median penumbra volumes 65 mL vs 80 mL. Core agreement was good (r=0.82; ICC=0.81) but showed systematic bias (RAPID - Brainomix: −4.9 mL) with wide 95% limits of agreement (−50.5 to +40.7 mL). Penumbra agreement was lower (r=0.77; ICC=0.74) with larger bias (−19.4 mL) and wide limits (−130.9 to +92.0 mL). Mismatch classification agreed in 91.5% (κ=0.52). Clinically significant core discrepancies occurred in 20.1%. Threshold-based reclassification at the DAWN (<50 mL) and DEFUSE-3 (<70 mL) volumetric cutoffs was discordant in 7.0% and 3.4% of patients, respectively.
Conclusion RAPID and Brainomix correlate overall but are not interchangeable; systematic biases and wide limits of agreement can alter mismatch-based eligibility. Rigid CTP thresholds for EVT selection should be applied cautiously and integrated with broader clinical assessment.
