Hyperdense Artery Sign in Patients With Acute Ischemic Stroke-Automated Detection With Artificial Intelligence-Driven...

Area of Science:

• Neuroradiology
• Artificial Intelligence in Medicine
• Stroke Imaging

Background:

• The hyperdense artery sign (HAS) on non-contrast computed tomography (NCCT) is a key indicator of large vessel occlusion (LVO) in acute ischemic stroke.
• Accurate HAS detection aids in identifying stroke patients who may benefit from specific interventions, even when LVO is not initially suspected.

Purpose of the Study:

• To evaluate the performance of a commercial AI-driven software for automated HAS detection.
• To compare the automated HAS detection accuracy against that of trained neuroradiologists.
• To assess the software's ability to automatically estimate thrombus length.

Main Methods:

• NCCT scans from 154 patients (with or without LVO confirmed by CT angiography) were analyzed.
• Two neuroradiologists and an AI algorithm (Brainomix®) independently assessed for HAS.
• Sensitivity, specificity, and automated thrombus length estimation were compared to a reference standard.

Main Results:

• The AI software achieved a sensitivity of 0.77 and specificity of 0.87 for HAS detection.
• Human readers demonstrated sensitivities of 0.80 and 0.93, and specificities of 0.97 and 0.71.
• Automated thrombus length estimation showed moderate agreement (ICC 0.73) with the reference standard.

Conclusions:

• Automated HAS detection on NCCT using the tested software is feasible.
• The software's performance in detecting HAS and estimating thrombus length is comparable to that of experienced neuroradiologists.
• AI-powered tools show promise for improving efficiency and accuracy in acute stroke imaging interpretation.