Task-Based Assessment of Pediatric Fluoroscopy by Use of Vesico-Ureteral Reflux Grading
D Olivera Velarde*, J Cruz Bastida, N Baughan, E Marshall, K Blunt, K Feinstein, I Reiser, The University of Chicago, Chicago, IL
Presentations
TU-IePD-TRACK 1-6 (Tuesday, 7/27/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]
Purpose: To propose a methodology for task-based evaluation of pediatric fluoroscopy systems analogous to the clinical diagnostic task of grading the severity of vesicoureteral reflux (VUR) in neonates.
Methods: Four depictions of VUR corresponding to grades 2-5 were created based on published references and embedded into selected regions from 20 neonate radiographs. Synthetic patches were simulated to match the noise-power spectrum of the selected regions, with mean and standard deviation matched to each radiograph region. A reader study program was written in MATLAB to display a simulated background image with an embedded task and a random displacement and rotation of the task of up to 15 pixels and 20-degrees, respectively. The reader was then prompted to assign the task a grade, from 2-5. Three observers performed a total of 15 reader studies using the simulated patches and five different contrast levels, determined as number of printing passes. Accuracy of grade identification for each contrast level and sensitivity of each class were calculated, with standard errors obtained using a bootstrapping technique.
Results: Accuracy increased as a function of contrast level with values of 0.45±0.054, 0.68 ±0.040, 0.70±0.020, 0.76±0.035, and 0.95±0.0094 for 10, 12, 13, 15, and 17 passes, respectively. In this study guessing would yield an accuracy of 0.25. Grades four and five were more easily confounded yielding sensitivity values of 0.68± 0.040 and 0.66±0.044, respectively. Grades two and three were easier to identify, and yielded sensitivity values of 0.79±0.040 and 0.78± 0.037.
Conclusion: The proposed methodology was used to simulate a diagnostic task and to establish the feasibility of using VUR grading as a surrogate task for quantitative, task-based assessment of the image quality of pediatric fluoroscopy. Future work will refine the reader study to improve reproducibility, and will extend this methodology to image quality assessment of real fluoroscopy systems.
Funding Support, Disclosures, and Conflict of Interest: This project was supported by the Department of Radiology Hodges Society, the Women's Board of the University of Chicago, the NIH T-32 Training Grant T32 EB002103, and the 2020-2022 AAPM Graduate Fellowship.
ePosters
Keywords
Taxonomy
Contact Email