Pioneering artificial intelligence-based real time assistance for intracranial liquid embolization in humans: an initial experience.

Abstract

Background: Liquid embolization in neuroendovascular procedures carries the risk of embolizing an inappropriate vessel. Operators must pay close attention to multiple vessels during the procedure to avoid ischemic complications. We report our experience with real time artificial intelligence (AI) assisted liquid embolization and evaluate its performance.

Methods: An AI-based system (Neuro-Vascular Assist, iMed technologies, Tokyo, Japan) was used in eight endovascular liquid embolization procedures in two institutions. The software automatically detects liquid embolic agent on biplane fluoroscopy images in real time and notifies operators when the agent reaches a predefined area. Safety, efficacy, and accuracy of the notifications were evaluated using recorded videos.

Results: Onyx or n-butyl-2-cyanoacrylate (NBCA) was used in the treatment of arteriovenous malformation, dural arteriovenous fistula, meningioma, and chronic subdural hematoma. The mean number of true positive and false negative notifications per case was 31.8 and 2.8, respectively. No false positive notifications occurred. The precision and recall of the notifications were 100% and 92.0%, respectively. In 28.3% of the true positive notifications, the operator immediately paused agent injection after receiving the notification, which demonstrates the potential effectiveness of the AI-based system. No adverse events were associated with the notifications.

Conclusions: To the best of our knowledge, this is the first report of real time AI assistance with liquid embolization procedures in humans. The system demonstrated high notification accuracy, safety, and potential clinical usefulness in liquid embolization procedures. Further research is warranted to validate its impact on clinical outcomes. AI-based real time surgical support has the potential to advance neuroendovascular treatment.