2023 ARRS ANNUAL MEETING - ABSTRACTS

RETURN TO ABSTRACT LISTING


1928. Artificial Intelligence Reduces Turnaround Time For Interpretation of CT Angiograms For Acute Pulmonary Embolism by Active Reprioritization
Authors * Denotes Presenting Author
  1. Kiran Batra *; UT Southwestern Medical Center
  2. Yin Xi; UT Southwestern Medical Center
  3. Ronald Peshock; UT Southwestern Medical Center
Objective:
To study the effect of artificial intelligence (AI)-driven worklist reprioritization on the turnaround time for CT pulmonary angiogram examinations for acute pulmonary embolism (PE) in a time of ever-increasing imaging volumes and radiology staff shortage.

Materials and Methods:
This was a retrospective study performed at a university hospital. Informed consent was waived. CT angiogram (CTA) examinations of the pulmonary arteries performed between September 1, 2018 and March 30, 2020 were evaluated. The study period was divided into phase 0 prior to the integration of the AI tool and phase 1 following the integration of the AI tool in the workflow. In phase 0, the examinations were prioritized from high to low on the basis of the clinical order as stat, urgent and routine. In phase 1, dedicated CTA examinations of the pulmonary arteries were routed for evaluation by an AI tool (AIDoc, New York, NY) for acute PE. The result of the AI was then returned to the PACS worklist (Primordial and iSite, Philips Medical Systems). Examinations flagged as positive for acute PE by the AI were reprioritized to the top of the worklist. Process timestamps were extracted from the EMR (Epic Radiant, Verona, WI) and the dictation system (Nuance Powerscribe, Burlington, MA). End-of-examination time was defined as the time stamp corresponding with the technologist marking the examination ended in the EMR. Wait time (WT) was defined as lenghth of time between when the exam was opened for reading in the dictation system, and when the study was marked complete in the EMR). Reading time (RT) = (Time when the first report [preliminary or final] is sent to the EMR) - (Time the exam is opened for dictation). Report turnaround time (RTAT) = WT + RT. Statistical analysis was used to assess the impact of the reprioritization on RTAT, WT, and RT, a generalized linear model with negative binomial distribution was used. A step-down Bonferroni adjustment was applied to p values when performing multiple comparisons.

Results:
1751 examinations, including 1191 from phase 0 and 560 from phase 1 were evaluated. RTAT for AI PE positive examinations was significantly decreased for routine but not urgent and STAT examinations (Mean TAT routine, phase 1: 45 min (95% CI: 31 – 65 min) vs. Mean RTAT routine examinations, phase 0: 86 min (95% CI: 77 - 95 min), p =0.01. RTAT for AI PE positive examination was also decreased compared to nonreprioritized, routine true PE negative examinations in phase 1 (96 min (95% CI: 80 – 95 min). The mean WT for routine AI PE positive examinations in phase 1 was 13 min (95% CI: 7 min – 25 min), as compared to the mean WT of routine examinations in phase 0 (59 min 95% CI 48 min – 71 min, p = 0.0017), and mean WT of the non-prioritized routine examinations in phase 1 (68 min, 95% CI 47 min – 97 min, p < .0001). No significant difference was observed for any other comparisons.

Conclusion:
AI-driven worklist reprioritization reduces the turnaround time of routine priority, CT angiography examination reports in patients with acute pulmonary embolism, thereby expediating management of these patients