Machine Learning Automated Detection of Large Vessel Occlusion From Mobile Stroke Unit Computed Tomography...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Prehospital detection of large vessel occlusions (LVO) in acute ischemic stroke patients via Mobile Stroke Units (MSUs) can expedite treatment.
• This study evaluates a machine learning (ML) model's efficacy in detecting LVO on CT angiograms (CTAs) acquired in an out-of-hospital setting.

Purpose of the Study:

• To assess the performance of an ML model in identifying LVO using CTAs from MSUs.
• To determine if prehospital LVO detection can be achieved accurately and rapidly.

Main Methods:

• A ML model was trained on in-hospital CTAs and subsequently tested on out-of-hospital CTAs from two MSUs.
• LVO was defined as occlusion in anterior circulation vessels and confirmed by expert readers.
• Model performance was quantified using the area under the receiver-operator curve (AUC).

Main Results:

• The ML model achieved an AUC of 0.84 on an in-hospital dataset and 0.80 on MSU CTA images.
• Analysis time for the ML algorithm was less than one minute.
• Forty percent of the 68 patients evaluated on MSUs had an LVO, most commonly in the middle cerebral artery M1 segment.

Conclusions:

• A machine learning algorithm demonstrated accurate and rapid detection of LVO using prehospital CTA data from MSUs.
• This technology holds promise for improving the speed of diagnosis and treatment initiation for stroke patients in prehospital settings.