Automated Intracranial Clot Detection: A Promising Tool for Vascular Occlusion Detection in Non-Enhanced CT

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Evaluating the diagnostic performance of a novel fully convolutional neural network (CNN)-based software prototype.
• Focusing on clot detection in intracranial arteries using non-enhanced computed tomography (NECT) imaging data.

Purpose of the Study:

• To assess the accuracy of an automated clot detection system in identifying intracranial arterial thromboembolism.
• To compare the software's performance against human expert readers and CT angiography (CTA) as ground truth.

Main Methods:

• Retrospective analysis of 85 patients with stroke and intracranial vessel occlusion.
• Automated computation of clot location, length, and volume using the NECT software prototype.
• Comparison of automated detection rates with visual assessment by neuroradiologists and CTA confirmation.

Main Results:

• The automated software achieved an overall clot detection rate of 66%, surpassing human readers (46% and 24%).
• Highest detection rates were observed in the internal carotid artery (ICA) and middle cerebral artery (MCA) (88-92%).
• A significant correlation was found between clot length and treatment choice (thrombectomy vs. conservative).

Conclusions:

• The automated clot detection prototype demonstrates potential for identifying intracranial clots in NECT scans, particularly in the ICA and MCA.
• This AI tool may assist radiologists in emergency settings, accelerating acute ischemic stroke diagnosis.
• The software offers a valuable alternative in situations where CTA is not readily accessible.