Endovascular thrombectomy is now recognized as the primary therapeutic approach for individuals presenting with ischaemic stroke caused by large vessel occlusion. Imaging software powered by artificial intelligence (AI) is gaining wider application to improve the detection and triage of suitable candidates for this procedure. This investigation examined the influence of AI stroke-imaging tools on the provision of endovascular therapy in England. We used a prospective observational design, drawing on information collected from stroke services across England’s National Health Service (NHS). Records for all patients aged 16 years or older admitted to NHS hospitals with a principal diagnosis of stroke were sourced from the national stroke registry, Sentinel Stroke National Audit Programme (SSNAP). We monitored endovascular thrombectomy utilization and interhospital transfer intervals through SSNAP at all 107 NHS hospitals handling acute stroke cases in England between 1 January 2019 and 31 December 2023. We assessed these indicators before and after the organized rollout of AI stroke software (Brainomix 360 Stroke) at 26 participating hospitals (six comprehensive stroke centers and 20 primary stroke centers; hereafter referred to as evaluation sites). We compiled aggregate hospital data for all institutions, while restricting detailed patient-level information to the evaluation sites. The principal measure was the percentage of stroke patients treated with endovascular thrombectomy. We compared treatment rate changes between individuals whose imaging was analyzed with AI assistance and those evaluated without such technology. During the study period from 1 January 2019 to 31 December 2023, 452 952 stroke patients were admitted to 107 hospitals. Patient-specific data were obtainable for 71 017 cases of ischaemic stroke admitted to the 26 evaluation sites. At these sites, endovascular thrombectomy was performed in 2.3% of patients (376 of 15 969) before implementation, rising to 4.6% (751 of 15 428) afterward—an exact doubling (100% relative increase). Non-evaluation sites showed an increase from 1.6% (1431 of 88 712) to 2.6% (2410 of 89 900), representing a 62.5% relative rise (odds ratio [OR] for site-by-time interaction 1.24 [95% CI: 1.08–1.43]; P = 0.0026). Patient-level analysis showed that AI software use was independently associated with higher odds of receiving endovascular thrombectomy (OR: 1.57 [95% CI: 1.33–1.86]; P < 0.0001) than standard care without AI support. To our knowledge, this investigation constitutes the largest prospective assessment of AI technology’s influence on clinical outcomes in stroke care. The results show a beneficial impact of AI imaging support on access to endovascular thrombectomy and reinforce current guideline recommendations to incorporate such software into routine stroke pathways.
