Kevin Koo, MD, MPH, on the accuracy of automated kidney stone detection on CT

In this video, Kevin Koo, MD, MPH, discusses findings from a recent study evaluating the accuracy and precision of automated kidney stone detection and measurement on CT scans using the quantitative Stone Analysis Software (qSAS), highlighting the potential role of automated image analysis in improving stone characterization and clinical decision-making.¹ Koo is a professor of urology at Mayo Clinic College of Medicine and Science and co-director of the multidisciplinary stone clinic at Mayo Clinic in Rochester, Minnesota.

Koo noted that CT remains the gold standard imaging modality for evaluating nephrolithiasis, providing critical information on stone size, location, density, and composition that influences patient counseling and treatment selection. To assess the performance of automated stone analysis, investigators conducted a phantom study using 120 kidney stones of known size and composition placed within an anthropomorphic phantom. Stones ranged from 1.4 to 9.9 mm and included calcium (n = 60), uric acid (n = 30), and mixed-composition calculi (n = 30). CT images were reconstructed using 1-, 3-, and 5-mm slice thicknesses and analyzed with qSAS to determine stone detection rates and maximum diameter measurements.

According to the study, qSAS demonstrated the highest accuracy on 1-mm CT slices, detecting 100% of stones, with only 5% of detected stones exhibiting a diameter error greater than 1 mm. Performance declined as slice thickness increased, with detection rates falling to 93% and 86% on 3- and 5-mm reconstructions, respectively. Diameter errors exceeding 1 mm occurred in 28% of stones on 3-mm slices and 64% on 5-mm slices. Investigators also observed composition-dependent measurement bias on thicker reconstructions, with uric acid stones tending to be underestimated and calcium-containing stones tending to be overestimated. Koo noted that larger stones, particularly those measuring at least 4 mm, remained reliably detectable even on thicker slices, whereas smaller and lower-density uric acid stones were more likely to be missed.

Koo noted that the findings have important implications for the clinical use of automated image analysis tools. In a complementary analysis of 45 kidney stones from 24 patients, agreement between qSAS and a radiologist was high when evaluating 1-mm CT images. However, measurement differences exceeding 1 mm were observed in 51% of stones, suggesting that automated assessments may offer greater precision than conventional manual measurements. Overall, the study demonstrates the potential of automated CT-based stone analysis to improve measurement consistency while also highlighting the continued influence of CT acquisition parameters and stone composition on diagnostic performance.

REFERENCE

1. Thongprayoon C, Ferrero A, Shi T, et al. Accuracy and precision of automated kidney stone detection on CT. Abdom Radiol (NY). 2026. doi:10.1007/s00261-026-05499-w