Estimating the impact of chest radiograph triage using AI – a real-life multicenter diagnostic cohort study.
Author
Plesner, Louis Lind | RegionH Denmark
Scientific poster presentation (W2-SPIN-3) at RSNA 2023, 26 – 30. November 2023 in Chicago, US
Purpose
Can an AI tool effectively triage CXR cases into remarkable and unremarkable categories in clinical practice?
Method
Retrospective validation of an AI model (Annalise Enterprise CXR), post-processed to provide unremarkable and remarkable distinction on 1.990 consecutive CXR studies with dual thoracic radiologist’s reference. The AI model was compared to binary classification extracted from RIS regarding various performance measures.
Results
The AI model demonstrated an AUC of 0.926 and was statistically superior to routine classification across all evaluated measures.
Key Takeaway
The AI model achieved excellent discrimination between unremarkable and remarkable CXRs and was superior to clinically assigned priority levels.
Disclaimer
Harrison.ai Radiology Solutions were previously marketed as Annalise.ai solutions.