A bicentric study on the application of dual energy CT for predicting hemorrhagic transformation post endovascular thrombectomy

Hemorrhagic transformation (HT) critically impacts outcomes in acute ischemic stroke (AIS) patients post-endovascular thrombectomy (EVT). Building upon the validated utility of post-EVT dual-energy CT (DECT), this study focused on developing and integrating a DECT-based predictive model with key clinical variables to achieve precise, individualized quantification of HT risk. This retrospective study analyzed 116 thrombectomy treated AIS patients stratified by HT status. Post-EVT DECT within 24 h assessed CT values (in Hounsfield Units, HU) of ischemic lesions on mixed energy images; CT values (in HU) on virtual non-contrast (VNC) images and on Sn80 keV and Sn150 keV monoenergetic images; absolute iodine concentrations (AIC, in mg/mL); and relative iodine concentrations (RIC, in %, where RIC = lesion AIC/sigmoid sinus AIC), using follow-up imaging and clinical criteria as the gold standard for HT.. HT patients exhibited higher NIHSS (median 14.5 vs. 9.0) and lower ASPECTS (9 vs. 13) than non-HT (nHT) counterparts, with elevated glucose (GLU, 8.26 vs. 6.45 mmol/L) and lower systolic blood pressure (SBP, 147.5 vs. 156.5 mmHg) (all P < 0.050). DECT-derived parameters demonstrated diagnostic utility, with both iodine overlay maps (IOM) and VNC positivity (χ² = 60.331, P < 0.001) and dual negativity (χ² = 58.870, P < 0.001) showing significant intergroup discrimination. Among 42 patients with IOM hyperdensity, RIC differed significantly between subgroups (t = - 2.566, P = 0.014), with elevated RIC independently associated with HT risk (OR = 1.040, 95% CI 1.003-1.078; P = 0.034). RIC alone exhibited strong predictive capacity for HT (AUC = 0.890, 95% CI 0.822-0.957). A nomogram-based model incorporating NIHSS, ASPECTS, and RIC achieved excellent HT prediction in both training (AUC = 0.947, 95% CI 0.903-0.991) and validation cohorts (AUC = 0.902, 95% CI 0.786-1.000), with stable calibration (training: P = 0.655; validation: P = 0.175) and clinical utility on decision curve analysis. Integration into stroke protocols may guide anticoagulation and secondary prevention decisions.