Abstract WMP67: Evaluation Of The Implementation Of An Ai Tool For Large Vessel Occlusion: Impact On Radiologists’ Workflow And Patient Outcomes

Abstract

Timely and accurate diagnosis of a large vessel occlusion (LVO) is critical for prompt initiation of thrombectomy treatment to reduce morbidity and mortality in acute ischemic stroke patients. Recently, artificial intelligence (AI) tools to aid stroke triage have shown promise for their accuracy and speed. However, little is known about the impact of the clinical integration of AI on efficiency metrics and patient outcomes. This study aims to evaluate the impact of an AI-based LVO detection tool on three main areas: (1) performance metrics, (2) radiologist workflow, and (3) ischemic stroke metrics and outcomes. This single center retrospective analysis includes patients with suspected acute stroke who underwent CT angiograms (CTA) in a three month period before and after the implementation of an AI-based LVO detection tool. CTA of 224 patients pre-implementation and 144 patients post-implementation of the tool were evaluated. The tool correctly identified 93/107 cases, with an accuracy of 87%, sensitivity of 91%, and specificity of 86%. The PPV was 0.64 and the NPV was 0.97. The mean radiology report turnaround time (TAT) using the tool was significantly reduced at 19.9 minutes (SD 32.4) compared to 33.7 minutes (SD 34.7) without the tool (p=0.001). There was a significant improvement in the resident trainee TAT with the LVO tool, with a mean TAT of 17.9 minutes (SD 32.6) compared to 38.4 minutes (SD 33.6) without (p=0.0002). Baseline stroke risk factors were similar between the pre and post-implementation groups. There were no significant differences in stroke metric medians for door-to-image, door-to-needle, door-to-puncture, and door-to-revascularization times and there were no statistically significant differences in NIHSS on arrival, post treatment NIHSS, discharge mRS, or mortality. The findings in clinical outcomes and stroke metrics may be attributable to small sample size. This study validates an AI-based tool in the detection of LVO and demonstrates a significant associated reduction in report TAT, allowing for faster communication of critical findings to ordering clinicians. Further research is needed to evaluate the impact of these tools on stroke benchmarks and clinical outcomes and to establish their role in stroke triage for clinicians.

Ähnliche Arbeiten