Multimodal deep learning using CT perfusion infarct core predicts post-thrombectomy parenchymal hematoma

Multimodal deep learning fusion model using CT perfusion-defined infarct core, clinical variables, and dual-phase CTA scores predicted parenchymal hematoma type 2 after endovascular therapy with AUC 0.845 in external validation (0.906 in training), outperforming clinical-imaging…

• Dual-center retrospective study of 487 anterior-circulation LVO acute ischemic stroke patients treated with EVT, split into training (n=219) and independent external validation (n=268) cohorts.
• The fusion model integrated infarct core NCCT patches, baseline clinical variables, and dual-phase CTA scores, achieving the highest performance.
• Interpretability analysis showed the model attended to low-density infarct core regions, and identified dual-phase CTA variables and baseline NIHSS as top predictors.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.