SIR 2024
Nonvascular Interventions
Meridith A. Kisting, MA (she/her/hers)
Research Assistant
University of Wisconsin Hospitals and Clinics
Financial relationships: Full list of relationships is listed on the CME information page.
Adrienne L. Kisting, BS
Research Assistant
University of Wisconsin
Financial relationships: Full list of relationships is listed on the CME information page.
Madeline Jentink, DO, DO
Resident Physician
UW Hospitals and Clinics
Disclosure information not submitted.
Giuseppe V. Toia, MD, MS
Assistant Professor
University of Wisconsin Hospital and Clinics
Disclosure information not submitted.
J. Louis Hinshaw, MD
Professor
University of Wisconsin School of Medicine and Public Health
Disclosure information not submitted.
Timothy J. Ziemlewicz, MD
Associate Professor of Radiology
University of Wisconsin Hospital and Clinics
Disclosure information not submitted.
Fred T. Lee, Jr., MD
Professor of Radiology, Biomedical Engineering, and Urology
University Of Wisconsin
Financial relationships: Full list of relationships is listed on the CME information page.
Martin G. Wagner, PhD
Assistant Professor
University of Wisconsin Madison
Disclosure information not submitted.
An AI algorithm was developed to suggest puncture pathways from volumetric lung CTs by segmenting thoracic anatomy and by detecting areas of emphysema. The algorithm scores over 50,000 pathways to a target based on literature-derived rules. To validate the algorithm, a reader study was performed with pathways generated from 28 volumetric lung CTs from patients who had previously received percutaneous lung biopsies (mean age = 68.4 ± 9.2, 12 female, 16 male). Five pathways were generated for each CT: the computer-selected best path and four random paths (n=140). Pathways were evaluated by four physicians (2-31 years of experience) and rated on a 1-5 scale, where 1-3 were safe and 4-5 were unsafe.
Results:
The correlation between computer and physician ratings was r = 0.86 (p < 1e-5). Pathways ratings were identical between the computer and physicians in 57.9% (81/140) cases and differed by a mean of 0.56 points. All 28 of the computer-selected best pathways were rated by physicians as safe and determined to be ideal or near-ideal. Many of the 28 best pathways utilized trajectories outside of the axial plane, with a mean gantry tilt of 9.1° ± 6.7°.
When binning ratings into safe (1-3) and unsafe (4-5), 5 pathways had discordant ratings of safe by computer and unsafe by physicians. After additional physician review, all 5 of these pathways were determined to be excessively long, with 2 pathways crossing vessels and one pathway crossing a rib and emphysema.
Conclusion: The AI algorithm can generate high-quality pathways and accurately judge the safety based on literature-derived rules. The computer-selected best pathways were determined by physicians to be ideal or near-ideal. Future integration with an anatomic guidance system would allow for needle placement at the precise location and angle suggested by the algorithm.