Fully Automated Longitudinal Assessment of Renal Stone Burden on Serial CT Imaging Using Deep Learning.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Kidney stone (urolithiasis) management requires accurate monitoring of stone burden.
• Serial CT scans are used to track changes in stone volume over time.
• Manual measurement of kidney stone burden can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for automated measurement and tracking of kidney stone burden.
• To assess the agreement between DL-based automated measurements and manual assessments of stone burden and its changes on serial CT scans.

Main Methods:

• A retrospective study included 259 CT scans from 113 patients with urolithiasis.
• A DL model was employed to detect, segment, and measure kidney stone volume (SV) on initial and follow-up scans.
• Automated measurements of stone burden (SV) and its changes (SVA, SVR) were compared to manual assessments using concordance correlation coefficient (CCC) and Bland-Altman plots.

Main Results:

• The DL pipeline achieved high per-scan sensitivity (97.8%) and positive predictive value (96.6%) for stone detection.
• Excellent agreement was found between automated and manual measurements for stone volume (CCC=0.995), absolute change (CCC=0.980), and relative change (CCC=0.915).
• The DL model effectively quantified stone burden and its interval changes, demonstrating high reliability.

Conclusions:

• Automated DL-based measurements provide accurate and reliable assessment of kidney stone burden on serial CT scans.
• DL technology offers a promising tool to streamline and improve the objectivity of urolithiasis monitoring.
• This approach has the potential to enhance clinical decision-making in the management of kidney stones.