Comprehensive analysis of 55,213 stones: understanding common morphological associations advances endoscopic stone recognition and AI integration.

PURPOSE

To assess the prevalence and associations of urinary stone morphologies, focusing on their relevance for Endoscopic Stone Recognition and improving AI-assisted ESR (AESR) systems.

METHODS

We analyzed a unique dataset comprising 55,213 stones classified by microscopic examination and Fourier transform infrared spectroscopy (FTIR) to assess the prevalence and associations of morphologies. We investigated the probabilities of observing a secondary morphology given a predominant morphology and the probabilities of observing a cross-sectional morphology given a surface morphology. Furthermore, we evaluated the performance of an AESR algorithm in detecting both pure and mixed stones across the six morphologies with a prevalence above 10% (Ia/Ib-Calcium oxalate monohydrate and IIa/IIb-Calcium oxalate dihydrate, IIIb-uric acid, and IVa-carbapatite), using a multi-label classification approach that integrates prior knowledge of common morphological associations.

RESULTS

This study provides key insights into common and rare urinary stone morphologies in a large cohort. The most frequent morphologies were Ia (55%), IVa (39%), and IIb (36%). We identified clinically relevant associations, notably Ia/Ib → IIb (28-30%), IIb → Ia (48%), and IIa → IVa (43%). We also showed that stone fragmentation often reveals deeper morphologies differing from the surface. In our experimental setup, AESR achieved mean balanced accuracies of 73% for surface images and 77% for section images, compared to 64% and 69% when morphological dependencies were ignored (Monte Carlo cross-validation over 10 random trials with 80% training, 10% validation, and 10% testing subsets).

CONCLUSIONS

Recognizing the prevalence and interrelationships of stone morphologies is essential for ESR. It improves both urologists' and AESR performance.

© 2025. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.