Presented by: Andrei D. Cumpanas MD
Department of Urology, University of California, Irvine

Introduction

Given the irregular shape of most renal stones, linear measurements, alone or as part of an ellipsoid formula , fail to accurately depict the true stone burden. We sought to train an AI algorithm to assess CT-based stone volume determinations.

Materials

A research physician fellow who was previously validated against a board-certified radiologist (interclass correlation coefficient of 0.99) established the “ground truth” for stone volume using 3D Slicer™ in 322 CT scans. To assess the convolutional neural networks’ (CNN) performance in determining stone volume both a Pearson correlation coefficient (R) and a Dice Overlap Score were calculated. Statistical analysis included a 5-fold cross validation. The volume of the stone with the largest linear measurement on a CT scan was determined by the UCI AI algorithm as well as by using the three best-fit ellipsoid formulas (Figure 1); these values were compared to the “ground truth”.

 

Results

,

The UCI AI algorithm was accurate(R=0.99) and precise (Dice score=0.96) for determining stone volume (Figure 2). The CNN outperformed the 3-ellipsoid formula-based volume predictions (Table 1). The algorithm’s accuracy and precision improved when measuring larger stones (i.e. > 2cm), as larger stones tended to have more irregular shapes; in contrast, the ellipsoid-determined volumes displayed an opposite trend (Table 1). Indeed,  even with the best of the ellipsoid formulas, the larger stone burden was overestimated by 27% to 88%.

Conclusion

The UCI Urology AI algorithm determined renal stone volumes in an accurate and precise way; it outperformed all three ellipsoid formulas.

Funding

None

Co-Authors

Chanon Chantaduly, BS
Center for Artificial Intelligence in Diagnostic Medicine, University of California, Irvine

Kalon L Morgan, MD
Department of Urology, University of California, Irvine;

Antonio R.H. Gorgen, MD
Department of Urology, University of California, Irvine;

Candices M. Tran, BS
Department of Urology, University of California, Irvine

Yi Xi Wu, PhD
Department of Urology, University of California, Irvine

Amanda McCormac, BS
Department of Urology, University of California, Irvine;

Sohrab N. Ali, MD
Department of Urology, University of California, Irvine;

Pengbo Jiang, MD
Department of Urology, University of California, Irvine;

Zachary E. Tano, MD
Department of Urology, University of California, Irvine;

Roshan M. Patel, MD
Department of Urology, University of California, Irvine;

Peter Chang, MD
Department of Radiological Sciences, University of California, Irvine;

Jaime Landman, MD
Department of Urology, University of California, Irvine;

Ralph V. Clayman, MD
Department of Urology, University of California, Irvine;