Smart chest X-ray worklist prioritization using artificial intelligence: a clinical workflow simulation

• DOI: 10.1007/s00330-020-07480-7
• Publication Date: 2020-11-21T16:02:36Z
• Authors: Ivo Baltruschat; Leonhard Steinmei...; Hannes Nickisch; Axel Saalbach; Michael Grass; Gerhard Adam; Tobias Knopp; Harald Ittrich
• Keywords: Prioritization

Abstract Objective The aim is to evaluate whether smart worklist prioritization by artificial intelligence (AI) can optimize the radiology workflow and reduce report turnaround times (RTATs) for critical findings in chest radiographs (CXRs). Furthermore, we investigate a method counteract effect of false negative predictions AI—resulting an extremely dangerously long RTAT, as CXRs are sorted end worklist. Methods We developed simulation framework that models current at university hospital incorporating hospital-specific CXR generation rates reporting pathology distribution. Using this, simulated standard processing “first-in, first-out” (FIFO) compared it with based on urgency. Examination was performed AI, classifying eight different pathological ranked descending order urgency: pneumothorax, pleural effusion, infiltrate, congestion, atelectasis, cardiomegaly, mass, foreign object. introduced upper limit maximum waiting time, after which highest urgency assigned examination. Results average RTAT all significantly reduced simulations FIFO (e.g., pneumothorax: 35.6 min vs. 80.1 min; p < 0.0001), while most increased same time 1293 vs 890 0.0001). Our “upper limit” substantially classes 979 min/1178 Conclusion demonstrate AI maintaining small FIFO. Key Points • Development realistic clinical simulator empirical data from allowed precise assessment using intelligence. Employing without threshold runs risk greatly increasing time. Use state-of-the-art convolution neural network almost perfect classification algorithm 30.4 min).