Malignant cerebral edema after endovascular thrombectomy: a multimodal prediction model based on…

AI / informaticsFrontiers in medicine· 2026

Malignant cerebral edema after endovascular thrombectomy: a multimodal prediction model based on post-thrombectomy cerebral hyperdensity and natural language processing.

Guolan Song, Jiahong Fu, Yuhan Chen … Jian Ding

NeuroendovascularAI / informaticsQ1IF 3.4

Multimodal AI predicts malignant cerebral edema after thrombectomy with 80% accuracy

Malignant cerebral edema after endovascular thrombectomy: a multimodal prediction model based on post-thrombectomy cerebral hyperdensity and natural language processing.

Guolan Song … Jian Ding

Frontiers in medicine · 2026 · Q1 in General Medicine

Background

Early prediction of malignant cerebral edema (malignant cerebral edema) following endovascular thrombectomy (endovascular thrombectomy) is critical for guiding timely interventions.

Purpose

This study aimed to develop and validate a multimodal prediction, integrating non-contrast CT (non-contrast CT) features and natural language processing (natural language processing)-encoded clinical data to predict malignant cerebral edema after endovascular thrombectomy.

Methods

In this multi-center retrospective study, 373 patients treated with endovascular thrombectomy were included, comprising internal ( n = 287) and external ( n = 86) cohorts.

n = 373 patients

Results

AUC of 1.0 is perfect prediction; 0.800 shows good accuracy identifying malignant edema risk

AUC 0.80 (95% CI 0.70 to 0.90)

null 0

0.70

0.90

CI excludes the null - significant

More results

SHapley Additive exPlanations (SHAP) analysis revealed natural language processing-derived semantic features as the principal predictors.

Furthermore, AI assistance improved the diagnostic performance of senior neuroradiologists (AUC: 0.709-0.763; p < 0.05) and increased their specificity increased from (78.1% to 84.4%).

“

Conclusion · 1 of 2

A multimodal framework integrating targeted non-contrast CT imaging features with natural language processing-encoded clinical data yields an accurate multimodal tool for early malignant cerebral edema prediction.

Conclusion · 2 of 2

This multimodal approach enhances human decision-making in emergency workflows.

Malignant cerebral edema after endovascular thrombectomy: a multimodal prediction model based on post-thrombectomy cerebral hyperdensity and natural language processing.

Malignant cerebral edema after endovascular thrombectomy: a multimodal prediction model based on post-thrombectomy cerebral hyperdensity and natural language processing.

Related papers