Machine learning integrating CT radiomics and clinical variables for preoperative prediction of renal uric acid stones: a model development and validation study

Introduction

Renal stone disease is a common urological condition, and its global prevalence has increased over recent decades. Chinese epidemiological data have shown a substantial rise in stone prevalence, from approximately 1.16% in the late 1970s to 7.54% by 2016 (1). International surveillance has revealed similar upward trends, with prevalence in the United States increasing from 0.6% in 2005 to 0.9% by 2015 (2), while German registry data showed an increase from 4% to 4.7% between 1997 and 2001 (3).

This increasing prevalence results from multiple interrelated factors, including genetic, environmental, and lifestyle influences, as well as metabolic disorders, dietary changes, and obesity. Stone disease also has a high recurrence rate, with approximately 50% of patients experiencing recurrence within 5–10 years and 75% within 20 years (4,5). Accurate determination of stone composition is essential for appropriate treatment selection and recurrence prevention. However, preoperative prediction of stone composition remains challenging, as definitive analysis traditionally requires surgically removed or spontaneously passed stones.

Current treatment strategies include both pharmacological and surgical interventions, with minimally invasive procedures increasingly being adopted. Uric acid stones are clinically important because they may respond to urinary alkalinization and pharmacological dissolution in appropriately selected patients (6). Therefore, accurate preoperative identification of uric acid stones may help guide individualized treatment planning and may reduce the need for surgical intervention in selected cases.

Current limitations in preoperative diagnostic capabilities present important clinical challenges. Although dual-energy computed tomography (DECT) and other advanced imaging modalities are effective for characterizing uric acid stones, their widespread use remains limited by high cost and restricted accessibility (7). Standard non-contrast computed tomography (CT) Hounsfield unit (HU) measurements can provide initial guidance regarding stone composition, but their predictive accuracy is often insufficient for definitive preoperative diagnosis (8).

Artificial intelligence (AI) and machine learning methods provide opportunities to analyze complex clinical and imaging data and may support non-invasive prediction of stone composition (9,10). Radiomics enables the extraction of quantitative features from medical images, including information related to intensity distribution, shape, texture, and spatial heterogeneity. Compared with conventional HU measurements, radiomics may capture additional imaging characteristics of stones that are not apparent through visual assessment or simple attenuation measurements (11). However, the added value of integrating CT radiomics features with routinely available clinical information for the specific prediction of renal uric acid stones remains insufficiently established.

The rationale for combining clinical information with CT radiomics is that these two data sources may provide complementary perspectives. Routine clinical information reflects patient-level metabolic and physiological conditions associated with stone formation, whereas CT radiomics describes quantitative imaging phenotypes of the stone itself. By integrating these complementary sources of information, a clinical-radiomics model may better characterize both the patient background and stone imaging phenotype than either source alone.

The persistent difficulty in reliably predicting stone composition before surgery remains an important challenge in the management of urolithiasis. In this study, we aimed to develop and validate machine learning models integrating standard non-contrast CT radiomics features with clinical variables for the preoperative identification of renal uric acid stones. We further compared several tree-based machine learning algorithms to explore their performance in radiomics-only and integrated clinical-radiomics prediction models. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0309/rc).

Methods

Study design and population

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Independent Ethics Committee (IEC) of The Fourth Affiliated Hospital of Soochow University (approval No. 241142, dated November 2024). The IEC waived the requirement for written informed consent owing to the retrospective design of the study and the use of anonymized data.

We conducted a retrospective prediction model development and validation study of 371 patients with kidney stones treated at The First Affiliated Hospital of Soochow University between May 2019 and July 2024. All calculi were surgically removed by percutaneous nephrolithotomy (PCNL), and stone composition was analyzed by infrared spectroscopy in our institutional urology laboratory. The study population included 240 male and 131 female patients. Patients were randomly assigned to training (n=296) and test (n=75) cohorts at an 8:2 ratio. According to compositional analysis, 60 patients had uric acid stones and 311 had non-uric acid stones.

Stone classification was based on the predominant component, with a single component accounting for ≥50% of the total composition used to define the stone type. For example, a specimen containing 70% anhydrous uric acid and 30% calcium oxalate monohydrate was classified as a uric acid stone.

Stone composition was determined by infrared spectroscopy according to routine laboratory procedures. No additional blinding procedure for outcome assessment was performed in this retrospective study. No formal a priori sample size calculation was performed because of the retrospective design. The final sample size was determined by the number of eligible patients treated during the study period. We acknowledge that the limited number of uric acid stone cases may increase the risk of overfitting, particularly given the high-dimensional radiomics feature space. To reduce this risk, dimensionality reduction was performed before model development, and model performance was evaluated using cross-validation and an internal held-out test set.

Patient selection criteria

Inclusion criteria

(I) Available preoperative abdominal non-contrast CT images; (II) stone removal by PCNL; and (III) postoperative stone composition analysis by infrared spectroscopy with clear classification as uric acid or non-uric acid.

Exclusion criteria

(I) Severe concomitant cardiovascular, neurological, or pulmonary disease; (II) presence of a nephrostomy tube, ureteral stent, or catheter during CT examination; (III) multiple stone specimens with different compositions from the same patient; (IV) stone diameter <5 mm; (V) ureteral, bladder, or other non-renal urinary stones; (VI) significant respiratory or motion artifacts on imaging; (VII) previous ipsilateral kidney lithotripsy; (VIII) use of uric acid-lowering or lipid-lowering medications; and (IX) incomplete clinical information.

Patients with incomplete clinical information were excluded before analysis; therefore, a complete-case analysis was performed, and no data imputation was applied.

Clinical data collection

Clinical data included demographic characteristics, including age, sex, body mass index, and medical history, including hypertension and diabetes and laboratory parameters, including serum uric acid, urinary pH, creatinine, triglycerides, total cholesterol, and HDL cholesterol. Stone composition data were obtained from the urology laboratory database. Because of the retrospective design, no formal blinding procedure for predictor assessment was performed. Clinical laboratory parameters were obtained as part of routine preoperative assessment before PCNL. Non-contrast CT images used for radiomics analysis were acquired before surgery.

Imaging analysis and region definition

CT examinations were performed using scanners from four manufacturers (Philips Brilliance 64, Siemens, GE, and Toshiba) according to standardized protocols: tube voltage, 100–120 kV; automatic tube current modulation; rotation time, 0.8 s; pitch, 0.984; and slice thickness, 1 mm. Preoperative non-contrast CT images were retrieved from the institutional Picture Archiving and Communication System (PACS). The images were carefully screened to exclude artifacts, ureteral stents, or concurrent non-renal stones. DICOM images were imported into Insight Segmentation and Registration Toolkit-Semi-Automatic Segmentation (ITK-SNAP) software, and the window width and level were adjusted to 350 HU and 50 HU.

Regions of interest (ROIs) were manually delineated slice-by-slice on axial sections to encompass the complete stone volume from the superior to inferior margins, supplemented by sagittal and coronal verification to ensure maximal stone inclusion while minimizing surrounding tissue contamination (renal parenchyma, vessels, fat, artifacts). Two experienced urologists collaboratively performed ROI segmentation with a third urologist providing review. Inter-observer reliability assessment using intra-class correlation coefficients (ICCs) demonstrated acceptable agreement (ICC >0.80), ensuring segmentation process reliability. The resulting ROIs and corresponding label files were stored in NIfTI format (.nii.gz) using anonymized study identifiers (Figure 1).

Figure 1 Region of interest using ITK-SNAP. (A) Single-layer region of interest outlining; (B) overall 3D model of the region of interest. 3D, three-dimensional; ITK-SNAP, Insight Segmentation and Registration Toolkit-Semi-Automatic Segmentation.

Radiomics feature analysis

Before feature extraction, CT images were resampled to a standardized isotropic voxel spacing of 1 mm × 1 mm × 1 mm to reduce variability related to acquisition parameters. Image intensity normalization was performed before radiomics feature extraction.

PyRadiomics software extracted radiomics features including: (I) first-order statistical measures; (II) shape-based characteristics; (III) gray-level co-occurrence matrix analysis; (IV) gray-level run-length matrix (GLRLM) evaluation; (V) gray-level size-zone matrix assessment; (VI) gray-level dependence matrix (GLDM) analysis; (VII) neighboring gray-tone difference matrix (NGTDM) computation. Features were derived from original images, five Laplacian of Gaussian (LoG) filters (σ=1.0–5.0 mm), and wavelet-transformed images (generating 8 sub-bands). Variance analysis and least absolute shrinkage and selection operator (LASSO) regression were used for dimensionality reduction. Only features with non-zero coefficients were retained for model development.

Computational model development

Decision tree (DT), random forest (RF), XGBoost, and CatBoost were used to construct prediction models because tree-based algorithms can model nonlinear relationships and potential interactions among predictors. Radiomics-only models were constructed using the selected radiomics features. Feature selection was performed using only the training cohort, and the selected feature set was then applied unchanged to the held-out test cohort. Integrated clinical-radiomics models were constructed by combining independent clinical predictors identified by logistic regression with the selected radiomics features. Model performance was evaluated using five-fold cross-validation and metrics including the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value.

Statistical analysis

Statistical analyses employed SPSS version 25.0 (IBM Corp., USA). Continuous variables are presented as mean ± standard deviation (SD) or median (interquartile range) depending on normality assessments. Independent samples t-tests, corrected t-tests, or Mann-Whitney U tests enabled group comparisons. Categorical variables were compared using the chi-square test. Independent predictors were identified using logistic regression, and DeLong’s test was used for receiver operating characteristic (ROC) curve comparisons. A two-sided P value <0.05 was considered statistically significant.

Results

Baseline characteristics comparison

A total of 371 patients were included in the analysis, including 60 patients with uric acid stones and 311 patients with non-uric acid stones. Comparisons between uric acid and non-uric acid stone groups revealed statistically significant differences (P<0.05) in age, urine pH, diabetes prevalence, serum uric acid, creatinine, triglycerides, and HDL cholesterol (Table 1). No significant differences emerged for gender, body mass index (BMI), hypertension, or other evaluated parameters.

Stone composition distribution

Among 371 analyzed stones, 60 (16%) represented uric acid compositions while 311 (84%) were non-uric acid types. The latter included calcium oxalate (235, 63%), apatite (70, 19%), calcium hydrogen phosphate dihydrate (3 cases), cystine (2 cases), and magnesium ammonium phosphate hexahydrate (1 case).

Clinical predictive factor analysis

Univariate logistic regression analysis identified age [odds ratio (OR) =1.060, 95% confidence interval (CI): 1.034–1.087, P<0.001], urine pH (OR =0.081, 95% CI: 0.041–0.164, P<0.001), serum uric acid (OR =1.008, 95% CI: 1.005–1.011, P<0.001), serum creatinine (OR =1.020, 95% CI: 1.008–1.032, P=0.001), diabetes (OR =2.106, 95% CI: 1.017–4.363, P=0.045) and high-density lipoprotein (HDL) cholesterol (OR =0.055, 95% CI: 0.015–0.206, P<0.001) as potential factors associated with uric acid stone formation (Table 2).

Subsequent multivariate logistic regression analysis demonstrated that older age (OR =1.063, 95% CI: 1.027–1.099, P<0.001), lower urine pH (OR =0.106, 95% CI: 0.047–0.235, P<0.001), higher serum uric acid (OR =1.005, 95% CI: 1.002–1.009, P=0.003), and lower HDL cholesterol levels (OR =0.041, 95% CI: 0.008–0.212, P<0.001) were independent risk factors for uric acid stone formation (Table 3).

Radiomics model development and assessment

Radiomic feature extraction using PyRadiomics generated a total of 1,316 features from original CT images, LoG filter-processed images (σ=1.0–5.0 mm), and wavelet-transformed datasets. After variance analysis and LASSO regression, 10 radiomic features were ultimately retained: five GLRLM features, two first-order features, two NGTDM features, and one GLDM feature (Table 4). Details of the LASSO regression are presented in Figure 2.

Figure 2 LASSO regression screening imaging features. (A) tuning parameters (α=4.977023564332114e-5); (B) distribution of imaging feature LASSO coefficients; (C) histogram of LASSO coefficients of 10 imaging features. 3D, three-dimensional; LASSO, least absolute shrinkage and selection operator; MSE, mean squared error.

Using these selected radiomic features, four machine learning classifiers—DT, RF, XGBoost, and CatBoost—were trained to predict uric acid stones. The dataset comprised 248 non-uric acid and 48 uric acid stone patients in the training cohort, and 63 non-uric acid and 12 uric acid stone patients in the test cohort. ROC curves of the four radiomics-only models in the training and test sets are presented in Figure S1, and the comparison is shown in Figure 3. Model stability was evaluated by five-fold cross-validation, and the corresponding ROC curves are presented in Figure S2.

Figure 3 Comparison of ROC curves of the four radiomics-only models in the training (A) and test (B) sets. AUC, area under the receiver operating characteristic curve; ROC, receiver operating characteristic.

Integrated model development and assessment

We further developed comprehensive predictive models by integrating the 10 selected radiomic features with four independent clinical risk factors (age, urine pH, serum uric acid, HDL cholesterol). Using the same machine learning classifiers (DT, RF, XGBoost, CatBoost) and data partitioning, we assessed model performance via five-fold cross-validation. ROC curves of the four integrated clinical-radiomics models in the training and test sets are presented in Figure S3, and the comparison is shown in Figure 4. Five-fold cross-validation ROC curves of the four integrated models are presented in Figure S4.

Figure 4 Comparison of ROC curves of the four integrated clinical-radiomics models in the training (A) and test (B) sets. AUC, area under the receiver operating characteristic curve; ROC, receiver operating characteristic.

Model performance comparison

All radiomics-only models showed high internal test-set discrimination. The RF classifier achieved the highest test-set AUC (0.983), closely followed by DT and XGBoost (both AUC =0.981) and CatBoost (AUC =0.980). Detailed performance metrics in the training and test cohorts are shown in Table 5. Five-fold cross-validation was used to assess model stability, and the corresponding ROC curves are provided in Figure S2.

In the integrated clinical-radiomics models, test-set discrimination was further improved. The CatBoost classifier exhibited the highest test-set AUC (0.991), followed by RF (0.987) and XGBoost (0.983), with DT performing comparatively lower (0.929). Detailed performance metrics in the training and test cohorts are shown in Table 6. Five-fold cross-validation was used to assess model stability, and the corresponding ROC curves are provided in Figure S4.

DeLong’s test indicated no significant differences in AUCs among the radiomics-only models in the training and test cohorts. For the integrated clinical-radiomics models, selected pairwise comparisons showed statistically significant differences in test-set AUCs. Given the limited number of uric acid stone cases in the test cohort, these pairwise comparisons should be interpreted cautiously.

Discussion

Our study showed that radiomics-based and integrated clinical-radiomics machine learning models achieved high internal performance for identifying renal uric acid stones. These findings suggest that CT radiomics combined with clinical variables may provide a basis for future non-invasive decision support. However, because the study was retrospective and internally validated only, the results should be interpreted cautiously.

Multivariate logistic regression analysis demonstrated that age, urine pH, serum uric acid, and HDL cholesterol were independently associated with renal uric acid stone formation. Previous large-scale international studies (12,13) and recent research (14) similarly indicated increasing uric acid stone prevalence with advancing age. This association may be explained by age-related declines in renal ammonia production, resulting in reduced urinary pH and increased urate supersaturation (15). Additionally, elderly populations frequently exhibit diabetes and insulin resistance, further reducing urine pH and facilitating uric acid stone formation (16).

Urine pH plays a critical role in uric acid stone formation. At pH values below 5.5, uric acid remains largely undissociated, whereas higher urine pH promotes urate ion formation. Elevated serum uric acid concentrations similarly increase urinary uric acid levels, promoting supersaturation and crystallization. Previous studies consistently reported low urine pH, elevated serum uric acid, and reduced urine volumes as major risk factors for uric acid stone formation, while obesity and hyperglycemia further enhance this risk (17-20). However, our study did not observe significant associations between uric acid stones and BMI or hypertension, possibly due to limited sample size.

Dyslipidemia has established connections to kidney stone formation. Previous research demonstrated elevated triglyceride levels in patients with uric acid stones, indicating that hypertriglyceridemia and decreased HDL cholesterol significantly increase stone risk (21). Dyslipidemia, particularly hypertriglyceridemia, can alter urine metabolic profiles, increasing calcium and uric acid excretion while decreasing urinary pH (22,23). Our findings are consistent with these previous reports, but we did not analyze urinary metabolic data. Future studies should incorporate urinary metabolic analyses to enhance the predictive power of preoperative stone composition models.

Pharmacological dissolution therapy and urinary alkalinization represent cornerstone medical management approaches for uric acid stones, demonstrating approximately 80% success rates with reduced healthcare costs and decreased patient morbidity compared to surgical interventions. Ong et al. (24) reported complete or partial dissolution in 75% of uric acid stones following oral alkalinization therapy. However, conservative management efficacy depends on accurate compositional prediction. Current predictive methods based solely on HU (HU <450), radiolucency, and urine pH <6 lack sufficient accuracy and specificity (25). Thus, there remains hesitance among clinicians to broadly apply dissolution therapy due to concerns about misdiagnosis and inappropriate management.

Advanced radiomics and machine learning approaches offer promising solutions for improved stone composition prediction. Ganesan et al. (26) distinguished uric acid from calcium stones by analyzing differences in attenuation patterns within stone cores and edges, achieving sensitivity and specificity above 89%. Spettel et al. (27) utilized combined HU measurements and urine pH to differentiate stone types, reporting sensitivities and specificities of 86% and 99.4%, respectively.

Emerging studies employing radiomics-based machine learning further enhance diagnostic precision. Tang et al. (28) identified calcium oxalate stones from CT-derived radiomics features with excellent predictive performance (accuracy ~88%). Similarly, Wang et al. (29) constructed a nomogram integrating CT radiomics and clinical variables, achieving AUC values of 0.878 and 0.867 for training and validation datasets, respectively. Additionally, machine learning models successfully differentiated infectious stones from non-infectious ones, significantly outperforming traditional logistic regression models (30).

Our study extracted 1,316 radiomics features from CT images, ultimately selecting 10 optimal features using variance analysis and LASSO regression. Radiomics-only models achieved high internal test-set discrimination, with AUC values ranging from 0.980 to 0.983. Five-fold cross-validation further supported the stability of both radiomics-only and integrated clinical-radiomics models, as shown in Figures S1-S4. The integrated CatBoost model achieved the highest held-out test-set AUC of 0.991, with an accuracy of 0.933, sensitivity of 0.917, and specificity of 0.937. These findings suggest its potential as a non-invasive decision-support tool for identifying patients who may be considered for early pharmacological intervention, although external validation is required before clinical use.

Because a prediction model for uric acid stones may influence whether a patient is considered for dissolution therapy rather than immediate surgical intervention, both false-negative and false-positive predictions have clinical consequences. High sensitivity is desirable to avoid missing patients who may benefit from non-surgical treatment, whereas high specificity is important to reduce inappropriate conservative management in patients with non-uric acid stones. In this exploratory study, balanced sensitivity and specificity approaching or exceeding 0.85–0.90 were considered potentially clinically promising. However, calibration and clinical net benefit should be evaluated in future external validation studies. The CatBoost integrated model achieved high internal test-set performance, with a sensitivity of 91.7% and specificity of 93.7%. However, the test set included only a small number of uric acid stone cases, and the CIs around these estimates are expected to be wide. Therefore, these results should be interpreted cautiously and require confirmation in external prospective cohorts before clinical adoption.

Several limitations require acknowledgment. First, this was a retrospective single-center study, and both the training and test cohorts were derived from the same source population. Therefore, all reported performance metrics represent internal validation estimates only. The absence of external validation is the most important limitation of this study, and the model’s generalizability to other institutions, scanner protocols, reconstruction settings, and patient populations remains unknown. Second, the number of uric acid stone cases was limited, and the class distribution was imbalanced. Although dimensionality reduction and internal validation were performed, overfitting remains possible, particularly given the high-dimensional radiomics feature space. The very high AUC values observed in the internal test set should therefore be interpreted cautiously. Third, calibration and clinical utility analyses were limited. Future studies should evaluate model discrimination, calibration, and clinical net benefit in independent external cohorts. Fourth, ROI segmentation was performed manually, which is labor-intensive and may limit clinical applicability. Automated or semi-automated segmentation methods should be investigated in future studies.

Conclusions

This retrospective single-center study developed and internally validated machine learning models integrating CT radiomics features with clinical variables for the preoperative prediction of renal uric acid stones. Serum uric acid, age, HDL cholesterol, and urine pH were identified as independent predictors. The integrated CatBoost model showed the highest test-set discrimination among the evaluated models. This non-invasive approach may have potential as a decision-support tool for early uric acid stone identification and targeted medical management. However, external prospective validation in multicenter cohorts is required before clinical implementation.

Acknowledgments

The authors thank the clinical and laboratory staff involved in data collection and stone composition analysis.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0309/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0309/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0309/prf

Funding: This study was supported by the Suzhou Medical Innovation Research Project (No. SZM2023034).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0309/coif). All authors report that this study was supported by the Suzhou Medical Innovation Research Project (No. SZM2023034). The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Independent Ethics Committee (IEC) of The Fourth Affiliated Hospital of Soochow University (approval No. 241142, dated November 2024). The IEC waived the requirement for written informed consent owing to the retrospective design of the study and the use of anonymized data.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wang W, Fan J, Huang G, et al. Prevalence of kidney stones in mainland China: A systematic review. Sci Rep 2017;7:41630. [Crossref] [PubMed]
2. Tundo G, Vollstedt A, Meeks W, et al. Beyond Prevalence: Annual Cumulative Incidence of Kidney Stones in the United States. J Urol 2021;205:1704-9. [Crossref] [PubMed]
3. Hesse A, Brändle E, Wilbert D, et al. Study on the prevalence and incidence of urolithiasis in Germany comparing the years 1979 vs. 2000. Eur Urol 2003;44:709-13.
4. Kavoussi NL, Da Silva A, Floyd C, et al. Feasibility of stone recurrence risk stratification using the recurrence of kidney stone (ROKS) nomogram. Urolithiasis 2023;51:73. [Crossref] [PubMed]
5. Shpitzer SA, Shpunt I, Loebl N, et al. Risk stratification for repeat stone surgery: the role of stone composition. World J Urol 2025;43:203. [Crossref] [PubMed]
6. Abou Chakra M, Dellis AE, Papatsoris AG, et al. Established and recent developments in the pharmacological management of urolithiasis: an overview of the current treatment armamentarium. Expert Opin Pharmacother 2020;21:85-96. [Crossref] [PubMed]
7. McGrath TA, Frank RA, Schieda N, et al. Diagnostic accuracy of dual-energy computed tomography (DECT) to differentiate uric acid from non-uric acid calculi: systematic review and meta-analysis. Eur Radiol 2020;30:2791-801. [Crossref] [PubMed]
8. Rodríguez-Plata IT, Medina-Escobedo M, Basulto-Martínez M, et al. Implementation of a Technique Based on Hounsfield Units and Hounsfield Density to Determine Kidney Stone Composition. Tomography 2021;7:606-13. [Crossref] [PubMed]
9. Baxt WG, Skora J. Prospective validation of artificial neural network trained to identify acute myocardial infarction. Lancet 1996;347:12-5. [Crossref] [PubMed]
10. Cooper JG, West RM, Clamp SE, et al. Does computer-aided clinical decision support improve the management of acute abdominal pain? A systematic review. Emerg Med J 2011;28:553-7.
11. De Perrot T, Hofmeister J, Burgermeister S, et al. Differentiating kidney stones from phleboliths in unenhanced low-dose computed tomography using radiomics and machine learning. Eur Radiol 2019;29:4776-82. [Crossref] [PubMed]
12. Knoll T, Schubert AB, Fahlenkamp D, et al. Urolithiasis through the ages: data on more than 200,000 urinary stone analyses. J Urol 2011;185:1304-11. [Crossref] [PubMed]
13. Tsaturyan A, Bokova E, Bosshard P, et al. Oral chemolysis is an effective, non-invasive therapy for urinary stones suspected of uric acid content. Urolithiasis 2020;48:501-7. [Crossref] [PubMed]
14. Basiri A, Tahvildari A, Naji M, et al. Determination of the kidney stone composition using infrared spectroscopy in Iran at a national referral center during 2019-2023. Asian J Urol 2025;12:72-8. [Crossref] [PubMed]
15. Maalouf NM, Sakhaee K, Parks JH, et al. Association of urinary pH with body weight in nephrolithiasis. Kidney Int 2004;65:1422-5. [Crossref] [PubMed]
16. Ye Z, Zeng G, Yang H, et al. The status and characteristics of urinary stone composition in China. BJU Int 2020;125:801-9. [Crossref] [PubMed]
17. Shekarriz B, Stoller ML. Uric acid nephrolithiasis: current concepts and controversies. J Urol 2002;168:1307-14. [Crossref] [PubMed]
18. Sakhaee K. Epidemiology and clinical pathophysiology of uric acid kidney stones. J Nephrol 2014;27:241-5. [Crossref] [PubMed]
19. Tran TVM, Maalouf NM. Uric acid stone disease: lessons from recent human physiologic studies. Curr Opin Nephrol Hypertens 2020;29:407-13. [Crossref] [PubMed]
20. Cheng WY, Tseng JS. Urinary stone analysis and clinical characteristics of 496 patients in Taiwan. Sci Rep 2024;14:14115. [Crossref] [PubMed]
21. Ding Q, Ouyang J, Fan B, et al. Association between Dyslipidemia and Nephrolithiasis Risk in a Chinese Population. Urol Int 2019;103:156-65. [Crossref] [PubMed]
22. Kang HW, Seo SP, Kim WT, et al. Hypertriglyceridemia is associated with increased risk for stone recurrence in patients with urolithiasis. Urology 2014;84:766-71. [Crossref] [PubMed]
23. Ferraro PM, Taylor EN, Curhan GC. 24-Hour Urinary Chemistries and Kidney Stone Risk. Am J Kidney Dis 2024;84:164-9. [Crossref] [PubMed]
24. Ong A, Brown G, Tokas T, et al. Selection and Outcomes for Dissolution Therapy in Uric Acid Stones: A Systematic Review of Literature. Curr Urol Rep 2023;24:355-63. [Crossref] [PubMed]
25. Zheng J, Zhang J, Cai J, et al. Development of a radiomics model to discriminate ammonium urate stones from uric acid stones in vivo : A remedy for the diagnostic pitfall of dual-energy computed tomography. Chin Med J (Engl) 2024;137:1095-104. [Crossref] [PubMed]
26. Ganesan V, De S, Shkumat N, et al. Accurately Diagnosing Uric Acid Stones from Conventional Computerized Tomography Imaging: Development and Preliminary Assessment of a Pixel Mapping Software. J Urol 2018;199:487-94. [Crossref] [PubMed]
27. Spettel S, Shah P, Sekhar K, et al. Using Hounsfield unit measurement and urine parameters to predict uric acid stones. Urology 2013;82:22-6. [Crossref] [PubMed]
28. Tang L, Li W, Zeng X, et al. Value of artificial intelligence model based on unenhanced computed tomography of urinary tract for preoperative prediction of calcium oxalate monohydrate stones in vivo. Ann Transl Med 2021;9:1129. [Crossref] [PubMed]
29. Wang Z, Yang G, Wang X, et al. A combined model based on CT radiomics and clinical variables to predict uric acid calculi which have a good accuracy. Urolithiasis 2023;51:37. [Crossref] [PubMed]
30. Chen T, Zhang Y, Dou Q, et al. Machine Learning-Assisted Preoperative Diagnosis of Infection Stones in Urolithiasis Patients. J Endourol 2022;36:1091-8. [Crossref] [PubMed]