Machine Learning Aids Identification of Large Vessel Occlusions

Researchers at Rhode Island Hospital (Providence, USA) conducted a study to assess the performance of a deep learning (DL) algorithm in identifying LVOs on CTA studies. To do say, they trained the CNN with more than 7,000 exams for LVO, and tested it on a set that consisted of 10% of these same exams, which were classified into three categories: no LVO (83.5%); chronic LVO (13.6%); and acute LVO (2.9%). The researchers also used 683 positive LVO studies as an independent test set.

Using CTA, it is possible to demonstrate the anatomy of the aortic arch, carotid, and cerebral arteries, stenosis/thrombus of carotid or cerebral arteries, and an impression of the functioning collateral network. Lack of enhancement also provides an estimate of the cerebral blood flow reduction. The results of the study showed that the CNN had a 96% accuracy rate and a 5% false-positive rate for categorizing LVOs. The study was presented at the Radiological Society of North America (RSNA) 106th Scientific Assembly and Annual Meeting, held online during November 2020.

“Almost 800,000 people in the United States have a stroke each year, and about 10% of those are due to large-vessel occlusions. Time is brain, so getting patients treated sooner will help them retain more neurologic function,” said lead author and presenter Ian Pan, MD. “3D CNNs were effective in identifying series of interest and acute LVOs in CT angiography. At a five percent false-positive rate, over 40% of acute LVOs would be successfully identified without manual intervention.”

Advanced AI imaging analysis will enable first responders to bypass emergency departments and take patients directly to an endovascular catheterization laboratory, the operating room, a hospital stroke unit, or the neuro-intensive care unit to provide immediate treatment, including administration of tissue plasminogen activator (tPA) and the potent blood pressure drug nicardipine.

Related Links:
Rhode Island Hospital