Program Information
Variability of Dual-Energy CT Textural Features Across Patient Size, Radiation Dose and Scanner Model
A Ferrero*, C McCollough , Mayo Clinic, Rochester, MN
Presentations
MO-RAM-GePD-I-5 (Monday, July 31, 2017) 9:30 AM - 10:00 AM Room: Imaging ePoster Lounge
Purpose: In this investigation, we sought to determine how patient size, radiation dose and scanner model impact the accuracy of the prediction of renal stone fragility using dual-energy CT (DECT) in an in-vitro model.
Methods: 55 stones of known composition were embedded in a water phantom modeling an average patient size and scanned using two clinical DECT systems (SOMATOM Definition Flash and SOMATOM Force) with the routine stone protocols in our practice. Additionally, protocols using half the radiation exposure as well as a phantom size modeling a large patient were tested. Textural features describing internal and external morphology were automatically extracted for each stone. Stone fragility was measured by disintegrating each stone in a controlled ex vivo experiment using an ultrasonic lithotripter and recording the comminution time. Using the data from the average patient size phantom and the Force scanner at routine dose, a multivariable linear regression model was developed to predict time to comminution based on the measured textures. The model was then applied to each additional CT acquisition and residual errors between the predicted and measured stone fragility were computed.
Results: The best multivariate model in the reference dataset accounted for 71% of the variability in the measured stone fragility. The difference in RMSE between the reference protocol and the additional protocols investigated was less than 40%. Using a 3-class method that considers fragile stones with comminution time in the lower quartile and hard stones those in the upper quartile, no fragile stone is classified as hard and vice-versa, regardless of the CT acquisition protocol used.
Conclusion: A preliminary model to predict stone fragility using routine CT has been developed in vitro. The prediction accuracy of the model is not significantly affected by the scanner model, patient size or radiation dose used, facilitating future adoption in clinical practice.
Funding Support, Disclosures, and Conflict of Interest: The project described was supported by Grant number DK100227 from the National Institute of Diabetes and Digestive and Kidney Diseases. This work was also supported by the AAPM Research Seed Funding Initiative to Dr. Ferrero
Contact Email: