Advancements in Uric Acid Stone Detection: Integrating Deep Learning with CT Imaging and Clinical Assessments in the Upper Urinary Tract.

Abstract

Introduction: Among upper urinary tract stones, a significant proportion comprises uric acid stones. The aim of this study was to use machine learning techniques to analyze CT scans and blood and urine test data, with the aim of establishing multiple predictive models that can accurately identify uric acid stones.

Methods: We divided 276 patients with upper urinary tract stones into two groups: 48 with uric acid stones and 228 with other types, identified using Fourier-transform infrared spectroscopy. To distinguish the stone types, we created three types of deep learning models and extensively compared their classification performance.

Results: Among the three major types of models, considering accuracy, sensitivity, and recall, CLNC-LR, IMG-support vector machine (SVM), and FUS-SVM perform the best. The accuracy and F1 score for the three models were as follows: CLNC-LR (82.14%, 0.7813), IMG-SVM (89.29%, 0.89), and FUS-SVM (29.29%, 0.8818). The area under the curves for classes CLNC-LR, IMG-SVM, and FUS-SVM were 0.97, 0.96, and 0.99, respectively.

Conclusion: This study shows the feasibility of utilizing deep learning to assess whether urinary tract stones are uric acid stones through CT scans, blood, and urine tests. It can serve as a supplementary tool for traditional stone composition analysis, offering decision support for urologists and enhancing the effectiveness of diagnosis and treatment.