Pre-operatively predicting kidney stone recurrence: integrating radiomic features and clinical variables using...

Area of Science:

• Nephrology
• Medical Imaging
• Artificial Intelligence

Background:

• Radiomics and AI show promise in urinary stone research.
• The link between stone radiomics and recurrence is understudied.
• Predicting kidney stone recurrence pre-operatively is crucial.

Purpose of the Study:

• To develop a machine learning model for predicting kidney stone recurrence.
• To integrate radiomic features and clinical variables for enhanced prediction.
• To assess the model's performance in diverse patient cohorts.

Main Methods:

• A cohort of 540 kidney stone patients was used for training and internal testing.
• An external test set of 141 patients validated the model.
• Radiomic features from CT scans and clinical data were analyzed.
• A nomogram integrating clinical predictors and radiomic features was developed.

Main Results:

• The combined nomogram model achieved AUC values of 0.820 (training), 0.824 (internal test), and 0.786 (external test).
• The nomogram demonstrated superior predictive performance compared to clinical models alone.
• High-risk patients identified by the nomogram had significantly lower stone-free survival.
• The model maintained robust discriminative performance over 3, 5, and 7 years.

Conclusions:

• A nomogram combining clinical variables and radiomic features shows potential for predicting kidney stone recurrence.
• This tool can aid in assessing recurrence risk during patient follow-up.
• Further validation in larger, diverse populations is warranted.