Prospective Evaluation of Artificial Intelligence Imaging Support Software for Acute Ischemic Stroke in the Mayo Clinic Telestroke Network
Smith BH., Wolfe JG., Karam A., Demarkschalk BM., Hrdlicka CM., Nasr DM., Chukwudelunzu FE., Nord CA., Pahl EA., Fernandez C., Wood S., Woodhead ZV., Carone D., Harston G., English SW.
Objective: To explore the real-world impact of artificial intelligence-driven decision support imaging software for patients with acute ischemic stroke in a mature telestroke network in the United States. Patients and Methods: We conducted a prospective evaluation of stroke imaging support software in a robust, predominantly rural telestroke network (17 sites in Minnesota and Wisconsin). Data was collected from all patients who underwent video telestroke evaluation in a 3-month preimplementation period before installation of the software (from February 10, 2024 to May 9, 2024) and a 3-month postimplementation period while the software was in use (from May 10, 2024 to August 9, 2024). The preimplementation and postimplementation cohorts were directly compared (no control group included). Primary outcome measures were treatment rates and time to treatment (both treatment decision and delivery) for intravenous thrombolysis (IVT) and endovascular therapy (EVT); secondary outcomes included transfer rates, transfer times, and end user survey results. Results: Total of 444 telestroke cases were included in the preimplementation period, and 463 in the postimplementation period. Comparing preimplementation and postimplementation periods, the rate of IVT treatment delivery rose from 26.6% to 35.0% of potentially eligible patients (P=.24), whereas EVT treatment delivery remained at 31%. Time to IVT delivery reduced from 47 minutes to 41 minutes (P=.772), and time to EVT treatment rose from 156 minutes to 157 minutes (P=.771). Overall rates of treatment (IVT or EVT) rose from 23.1% to 23.9% of potentially eligible patients (P=.944). Although none of the clinical outcomes reached statistical significance, the survey results reported good user satisfaction with algorithm performance and image viewing. Conclusion: This study reported a nonsignificant increase in treatment rates and a decrease in time to treatment decisions. Future trials with larger sample sizes are needed to validate the real-world benefits of decision support software for acute ischemic stroke in an established telestroke network.