Implementation of an AI-supported imaging decision-making tool (Brainomix 360 Stroke) was associated with significant reductions in key workflow times, largely attributable to decreased times at primary stroke centres in the region.
Key Findings
Results
Adjusted median CT-to-groin-puncture time significantly decreased following implementation of Brainomix 360 Stroke.
Adjusted median CT-puncture time decreased from 47 minutes pre-implementation to 36 minutes in the learning period and 35 minutes in the established period.
The study included 970 EVT patients treated at Sahlgrenska University Hospital from June 2021 to May 2024.
The study period was divided into pre-implementation, learning, and established periods following Brainomix 360 Stroke introduction.
Decreases were statistically significant across the implementation periods.
Results
Adjusted median CT-to-perfusion-map-availability time significantly decreased from pre-implementation to the established period.
Adjusted median CT-to-perfusion-map-availability decreased from 7 minutes pre-implementation to 6 minutes in the established period.
This outcome measured the time from non-contrast brain CT (NCCT) to CT perfusion map availability.
The decrease was statistically significant.
Results
Workflow time improvements were largely attributable to changes at primary stroke centres (PSCs) rather than the comprehensive stroke centre.
Subgroup analyses of PSCs showed results similar to the overall cohort findings.
No significant differences were found in subgroup analyses of the comprehensive stroke centre.
This suggests the AI tool had differential impact depending on facility type.
Results
No significant differences in clinical outcomes were observed between study periods.
Clinical outcomes assessed included early neurological improvement, defined as a reduction of ≥4 points on the NIHSS or a score of 0–1 at 24 hours.
Favourable functional outcome was defined as mRS scores 0–2 or return to pre-stroke mRS at 90 days.
No significant differences in either clinical outcome measure were detected across the pre-implementation, learning, and established periods.
Methods
The study design was a retrospective register-based cohort study of consecutive EVT patients in a high-volume stroke system with routine perfusion imaging.
970 consecutive patients treated with EVT were included.
Patients outside the Region Västra Götaland prehospital area or aged under 18 years were excluded.
The study took place at Sahlgrenska University Hospital from June 2021 to May 2024.
The system already used routine perfusion imaging prior to AI tool implementation, representing a high-baseline workflow environment.
What This Means
This research studied whether adding an artificial intelligence (AI) tool called Brainomix 360 Stroke to a busy stroke treatment system could speed up the process of evaluating and treating patients who had a stroke caused by a large blocked blood vessel. The researchers looked at nearly 1,000 patients treated with a clot-removal procedure (endovascular thrombectomy) over three years and compared key time measurements before and after the AI tool was introduced.
The study found that after implementing the AI tool, two important time measurements improved significantly. The time from the initial brain scan to when doctors could puncture the groin artery to begin clot removal dropped from about 47 minutes to about 35–36 minutes. The time it took for perfusion imaging maps (which show how much brain tissue is at risk) to become available also shortened slightly. These improvements were mainly seen at smaller primary stroke centres in the region, while the large comprehensive stroke centre did not show the same gains — possibly because it was already operating efficiently. However, no significant improvements in patient outcomes, such as neurological recovery or functional independence at 90 days, were detected.
This research suggests that AI-assisted imaging tools can meaningfully reduce critical delays in stroke care workflows, particularly at smaller centres that may have less experience interpreting complex brain imaging. The lack of detectable improvement in patient outcomes may reflect the fact that the study was not designed with enough statistical power to detect such changes, or that other factors influence outcomes beyond imaging speed. These findings are relevant for hospital systems considering AI tools to streamline stroke care.
Karlsson A, Cadeborn E, Woock M, Allardt A, Jood K, Björkman-Burtscher I, et al.. (2026). Implementation of an AI-supported decision-making tool in a high-volume stroke system with routine perfusion imaging.. European stroke journal. https://doi.org/10.1093/esj/aakag098