ARRS 2022 Abstracts

RETURN TO ABSTRACT LISTING


1866. Artificial Intelligence Enhances Time Efficiency in Reading Chest CT Scans: A Randomized Prospective Study
Authors * Denotes Presenting Author
  1. Logan Fitzpatrick *; Medical University of South Carolina
  2. Rock Savage; Medical University of South Carolina
  3. Jim O'Doherty; Siemens Healthcare
  4. Tilman Emrich; Medical University of South Carolina
  5. Akos Varga-Szemes; Medical University of South Carolina
  6. U. Joseph Schoepf; Medical University of South Carolina
Objective:
This study aims to determine the impact of incorporating an artificial intelligence (AI) support platform into clinical radiology workflows on chest CT reading times for radiologists at an academic cardiothoracic practice.

Materials and Methods:
In this prospective study, reading times for chest CT scans by three board certified cardiothoracic radiologists were assessed. AI-Rad Companion Chest CT (Siemens Healthineers, Erlangen, Germany), an AI support platform was used, comprising modules for assessment of pulmonary hypodensity, pulmonary nodules, coronary calcifications, the thoracic aorta, and the thoracic spine. The availability of AI results on the institutional picture archiving and communication system, for review by the reading radiologists, was randomized by the CT technologists. Consecutive contrast and non-contrast CT scans were included in this study, and each reader was assigned an equal number of cases with and without AI results. Reading times were measured by the radiologists beginning after the cases had been loaded on the reporting workstations and ending prior to dictation. Cases that involved reading interruptions, e.g., phone calls, were excluded.

Results:
A total of 390 chest CT scans were included and allocated equally to the AI and non-AI assisted arms. Mean age of patients was 62.8 ± 13.3 years with 52.3% women. Mean reading time of the radiologists was shorter in the AI arm by 92.9 seconds (s) with a 95% CI: 62.9 - 123.0 s. This accounted for a mean reduction of reading time by 22.1% (95% CI: 14.9 - 29.2%) when compared to the mean non-AI reading time (AI: 328.2 ± 122.0 s; non-AI: 421.1 ± 175.0 s; p-value <0.001). This pattern of shorter mean reading times utilizing results from the AI support platform was observed on non-contrast CT scans (AI: 325.1 ± 100.7 s; non-AI: 428.7 ± 173.3 s; difference: 24.2%; p-value <0.001), and contrast-enhanced scans (AI: 331.0 ± 139.2 s; non-AI: 413.6 ± 177.1 s; difference: 20.0 %; p-value <0.001).

Conclusion:
There was a mean reduction of 22.1% in reading time of contrast-enhanced and non-contrast chest CT scans by cardiothoracic radiologists when results from an AI support platform were made available to the readers. The incorporation of AI support platforms into clinical radiology workflows may increase reading efficiency for radiologists and reduce turnaround times for reports.