Percutaneous Lung Biopsies Aided by Artificial Intelligence: A Comparison Between Computer and Physician-Chosen Biopsy Paths.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology, University of Wisconsin, Madison, WI.
- Department of Carle Illinois College of Medicine, University of Illinois Urbana-Champaign, Urbana, IL.
- Departments of Medical Physics.
- Biomedical Engineering.
- Urology, University of Wisconsin, Madison, WI.
Abstract
To compare Trajectory Recommendation Algorithm for CT-guided Biopsy (TRAX)-generated lung biopsy puncture pathways versus physician-chosen paths. TRAX is an artificial intelligence (AI)-based algorithm that uses segmentation and physician-chosen logic rules to generate lung biopsy pathways. Once a target lesion is defined by the physician, TRAX generates and ranks ∼20,000 candidate pathways within an axial angle of ±20°. Blinded radiologists retrospectively rated pathways chosen by physicians (n=53) versus TRAX (n=53) from the same patients and setup scans prior to lung biopsies (scale: 1 to 3 safe, 4 to 5 unsafe). The quality and metrics of the pathways were compared. All TRAX and physician-chosen pathways were determined safe by physician reviewers (rating 1 to 3). Ratings were identical in 93/159 (58%) cases; TRAX was superior in 36/159 (23%) cases, and physician paths were superior in 30/159 (19%) (no significant difference between pathways, P=0.61). TRAX pathways were shorter than physician pathways (7.2±2.5 vs. 7.8±2.1 cm, P=0.046). Most TRAX pathways were outside of the axial plane [n=27/53 (50.9%)], mean gantry angle=11.4±6.0°. The majority of physician-generated pathways were in the axial plane [n=43/53 (81.1%)], mean gantry angle=0.9±2.9° (TRAX vs. physician P<0.05 for proportion of paths in the axial plane and mean gantry angle). TRAX appears to be a promising AI tool to assist physicians in selecting needle trajectories for percutaneous CT-guided lung biopsies, particularly those outside the axial plane.