Spontaneous stone passage prediction in acute ureteric colic using radiomics and machine learning

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Urology

Background:

• Acute ureteric colic (AC) is a common urological emergency often managed based on predicting spontaneous stone passage (SSP).
• Current prediction methods may lack precision, necessitating advanced analytical tools.
• Non-contrast computed tomography (NCCT) is the standard for diagnosing ureteric stones.

Purpose of the Study:

• To develop and validate a machine learning (ML) model using radiomics features from NCCT to predict SSP in patients with uncomplicated AC.
• To assess the performance of ML models in predicting SSP based on CT-derived stone characteristics.

Main Methods:

• Retrospective analysis of 428 patients with uncomplicated AC confirmed by NCCT.
• Semi-automated segmentation of ureteric stones and extraction of radiomic features using PyRadiomics.
• Feature selection via LASSO, followed by training nine ML models, with Random Forest showing the best performance (AUC 0.79).

Main Results:

• The study included 428 patients with a median stone diameter of 5.3 mm; 47.9% achieved SSP.
• Seven selected radiomic features, including texture and intensity descriptors, were used for prediction.
• The Random Forest model achieved an AUC of 0.79, with 72% accuracy and 72% F1-score on the validation set.

Conclusions:

• Radiomics analysis of NCCT images can effectively predict SSP in uncomplicated AC.
• Integration of stone location into hybrid models may further enhance predictive accuracy.
• External validation and development of automated tools are recommended for clinical implementation.