The combination of artificial intelligence (AI) detection of large vessel occlusion (LVO) on computed tomography (CT) angiograms with subsequent real-time alerts to clinicians may significantly improve workflow and outcomes for the management of stroke patients.
For the randomized stepped-wedge clinical trial, recently published in JAMA Neurology, researchers reviewed data from a total of 243 patients (median age of 70) drawn from four comprehensive stroke centers (CSCs). All patients were treated for LVO with endovascular therapy (EVT). The study authors compared the use of AI LVO detection on CT (Viz.ai) and real-time alerts on the management of 103 patients versus 140 patients treated prior to the use of AI software.
The researchers found that the combination of the AI software and real-time alerts to the multidisciplinary team about possible LVO detection resulted in an 11.2-minute reduction in door-to-groin (DTG) time, the time between hospital arrival and the beginning of EVT. For patients in the post-AI intervention cohort, the study authors also noted a 9.8-minute reduction in the time from the beginning of the CT scan to the initiation of EVT.
Emphasizing the complexities of the multidisciplinary workup, assessment, and management of a patient with LVO with EVT, the researchers noted that potential LVO alerts from the AI software arrive via clinician mobile phones shortly after CT angiography has been completed. The study authors added that these alerts are often issued to the multidisciplinary team prior to availability of the CTA images in the PACS system.
“By incorporating all the relevant team members into a single communication platform, parallel processing may become easier. Our finding that time from CT initiation to EVT start fell by approximately 10 minutes, which is most of the time reduction seen in DTG, supports this possibility,” wrote study co-author Sunil A. Sheth, M.D., an associate professor and director of the Vascular Neurology Program in the Department of Neurology at McGovern Medical School at UTHealth in Houston, and colleagues.
Three Key Takeaways
1. Reduced triage time. The use of AI software for detecting large vessel occlusion (LVO) on CT scans resulted in an 11.2-minute reduction in door-to-groin (DTG) time for acute stroke patients, which can be critical in the initiation of endovascular therapy (EVT).
2. Improved patient outcomes. The combination of AI detection and real-time alerts led to a nearly 60 percent decrease in mortality rates among patients with possible LVOs, compared to those without AI assistance. This suggests that AI can positively impact patient outcomes in stroke management.
3. Applicability to specific patient groups. The study showed significant improvements in DTG time for patients with specific characteristics, such as those with higher NIHSS scores, anterior circulation LVOs, early presentations within six hours of last known well, and those not receiving intravenous tissue plasminogen activator (IV tPA). This highlights the potential benefits of AI software in streamlining care for certain stroke patient subgroups.
Specifically, sensitivity analysis showed DTG improvements in patients with a National Institutes of Health Stroke Scale (NIHSS) score greater than 10 (12.3 minutes); patients with anterior circulation LVOs (12.9 minutes); patient presentations within first six hours of last known well (13 minutes); and patients who didn’t receive an intravenous tissue plasminogen activator (IV tPA) (14 minutes).
The researchers also pointed out a nearly 60 percent decrease in mortality rates with the use of the AI software and real-time alerts of possible LVOs (12.6 percent) in comparison to patients in the non-AI cohort (31.4 percent).
(Editor’s note: For related articles, see “Study Shows Significant Overutilization of Head and Neck CT Angiography in the ER.” “Should Dual-Source CT be the New Standard for ER CCTA Assessment of Acute Chest Pain?” and “Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients.”)
Sheth and colleagues noted no differences between the AI and non-AI cohorts with respect to intracerebral hemorrhage (ICH), hospital length of stay and 90-day functional outcomes.
In regard to study limitations, while it was a multicenter study, all four CSCs in the study were part of the same health-care system and some physicians rotated between these centers. The researchers noted this consistency and similar workflows for the management of stroke imaging and care may limit extrapolation to stroke imaging and care at other facilities.
