Article Text

Download PDFPDF

E-249 Endovascular brain aneurysm occlusion using the web device: a machine learning approach to predict postprocedural immediate complete occlusion with a focus on ruptured aneurysms
Free
  1. M Essibayi1,
  2. M Jabal2,
  3. N Adeeb3,
  4. A Dmytriw4,
  5. D Altschul1
  1. 1Neurosurgery, Albert Einstein College of Medicine, Bronx, NY
  2. 2Radiology, Mayo Clinic, Rochester, MN
  3. 3Neurosurgery, Louisiana State University, Shreveport, LA
  4. 4Radiology, Harvard University, Boston, MA

Abstract

Background The Woven EndoBridge (WEB) device is a novel intrasaccular aneurysm treatment, but variable complete occlusion rates have been reported. Immediate occlusion is critical, especially in ruptured aneurysms, however determinants of successful WEB occlusion are undefined. This study aimed to identify factors associated with effective WEB occlusion outcomes, with a focus on ruptured aneurysms.

Methods A multicenter retrospective study compiled data on patients undergoing WEB treatment for intracranial saccular aneurysms from 2011–2022. Key demographic, clinical, morphological, and procedural variables were extracted. Patients were categorized into incomplete and complete immediate occlusion groups. Complete occlusion was defined as Raymond-Roy occlusion classification (RROC) score I while incomplete occlusion was defined as RROC II-III. Comparative analyses were conducted to identify differing characteristics between groups. Machine learning models were developed using relevant features to predict immediate occlusion.

Results A total of 1565 aneurysm patients treated with WEB devices were included, 1028 (65.7%) showed immediate incomplete, and 537 (34.3%) showed immediate complete occlusion after WEB deployment within the aneurysm sac. Comparative analyses found older age, smoking, subarachnoid hemorrhage history, larger/complex aneurysm morphology, and bifurcation locations were associated with incomplete occlusion. The CatBoost classifier model performed best for predicting incomplete occlusion, which achieved an AUC of 0.73. Key predictors included aneurysm neck diameter, middle cerebral artery location, smoking status, maximal diameter, multiple aneurysms, width, branches, access route, daughter sacs, and bifurcation site. Among 435 ruptured aneurysms, the CatBoost classifier model performed best for predicting incomplete occlusion, which achieved an AUC of 0.69. Overall, 269 (61.7%) demonstrated immediate incomplete occlusion, and 167 (38.3%) demonstrated immediate complete occlusion. Predictors of incomplete occlusion included pretreatment disability, aneurysm width, smoking history, morphological factors, and bifurcation site.

Conclusions This large real-world study identified diverse demographic, clinical, morphological, and procedural factors associated with WEB occlusion outcomes. A machine learning model enabled personalized risk stratification to optimize intracranial aneurysm treatment decision-making. Elucidating ruptured aneurysm-specific predictors provides vital insights to improve occlusion in this high-risk group. Further research is warranted to validate the occlusion prediction model across diverse populations and assess durability.

Disclosures M. Essibayi: None. M. Jabal: None. N. Adeeb: None. A. Dmytriw: None. D. Altschul: 1; C; Bee foundation. 2; C; Microvention.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.