Comparison of actual and automated CT measurements of urinary stone size: a phantom study

Abstract

Urinary stone size is key in determining treatment. Although computed tomography (CT) scans are widely used for diagnosing urinary stones, measurements of stone size obtained from CT images may be inaccurate compared to actual size. Twenty-four urinary stone phantoms were 3D printed at three densities (100, 1000, and 3000 Hounsfield units [HU]) and eight sizes. CT images of the phantoms were taken. Nineteen radiologists and 33 emergency physicians from two institutions measured stone sizes on CT images using mediastinum and bone settings. An automated algorithm segmented regions of interest and estimated stone size using pixel HUs. Mean absolute error (MAE) was assessed for the accuracy of each measurement method against known phantom sizes. For the mediastinum setting, MAEs for 100, 1000, and 3000 HU stone phantoms were 1.05 mm +/- 0.06, 1.01 mm +/- 0.06, and 2.38 mm +/- 0.17, respectively. For the bone setting, MAEs were 0.98 mm +/- 0.07, 0.55 mm +/- 0.10, and 1.91 mm +/- 0.06, respectively. For automated measurements, MAEs were 1.16 mm, 0.21 mm, and 2.10 mm, respectively. Participant-to-participant variability was observed across all measurement settings, regardless of the stone density or window used. For stone size measurements on CT images, the bone setting provided more accurate results than the mediastinum setting. Automated measurement methods, which estimate stone size by outlining its edges, were more accurate than manual measurements for 1000 HU stones, the most common stone density. However, for stones with densities above or below 1000 HU, the accuracy of the automated method may decrease.