Estimation of urinary stone composition by automated processing of CT images

TL;DR

This study developed an automated CT image analysis tool to estimate urinary stone composition, achieving comparable accuracy across different dose protocols and eliminating manual radiologist intervention.

Contribution

The paper introduces a fully automated method for urinary stone composition estimation from CT images, reducing manual effort and maintaining accuracy across various dose protocols.

Findings

Stone composition determined in 52% of cases.

Sensitivity for uric acid: 65%.

Sensitivity for cystine: 78%.

Abstract

The objective of this article was developing an automated tool for routine clinical practice to estimate urinary stone composition from CT images based on the density of all constituent voxels. A total of 118 stones for which the composition had been determined by infrared spectroscopy were placed in a helical CT scanner. A standard acquisition, low-dose and high-dose acquisitions were performed. All voxels constituting each stone were automatically selected. A dissimilarity index evaluating variations of density around each voxel was created in order to minimize partial volume effects: stone composition was established on the basis of voxel density of homogeneous zones. Stone composition was determined in 52% of cases. Sensitivities for each compound were: uric acid: 65%, struvite: 19%, cystine: 78%, carbapatite: 33.5%, calcium oxalate dihydrate: 57%, calcium oxalate monohydrate: 66.5%, brushite: 75%. Low-dose acquisition did not lower the performances (P < 0.05). This entirely automated approach eliminates manual intervention on the images by the radiologist while providing identical performances including for low-dose protocols.