Background: Thromboembolic events are a known risk of neuroangiographic procedures, and prompt
detection depends on the proceduralist’s advanced knowledge of cerebral vascular anatomy
and the dynamic timing of contrast transit. With ongoing advances in surgical adjuncts
and software, subtask automation with artificial intelligence (AI) models helps to
reduce technical errors and make neurosurgical procedures safer. This study serves
as proof-of-concept for a preliminary AI model for automatic thrombus recognition
during an angiographic procedure.
Methods: An automated image processing and visualization pipeline was developed for angiographic
image analysis using the DICOM files. First, faulty angiographic images were eliminated
using a deep learning classifier, and then thresholding and stacking vessel maps were
created. At the final stage of imaging processing, pre- and post-thrombectomy vessel
maps were registered for visualization and clot detection.
Results: Eight vessel maps were created using the above-stated imaging processing techniques
to identify a vessel occlusion on angiographic imaging.
Conclusion: AI-based software algorithms present an enormous underdeveloped opportunity to improve
prompt identification of thromboembolic events during neuroangiographic procedures,
which offers a greater safety benefit to the patient. This study presents a preliminary
AI model for clot detection, which required significant manual imaging processing—a
model to build on for more robust AI-based image processing and clot detection in
the future. Ultimately, this software is intended to work in synergy with the neurointerventionalist
and current intraoperative monitoring practices to enhance simple and complex neuroangiographic
procedural outcomes.
