Predicting Extracorporeal Shock Wave Lithotripsy Outcomes Using Machine Learning and the Triple-/Quadruple-D Scores

By Mucahit Gelmis, Sina Kardas, Ali Ayten, Oguzhan Cura, Serkan Gonultas, Mustafa Gokhan Kose

Affiliations

1. Department of Urology, Gaziosmanpasa Training and Research Hospital, Istanbul, Turkiye

doi: 10.29271/jcpsp.2025.08.1007

ABSTRACT
Objective: To evaluate the predictive performance of the triple-D and quadruple-D scores integrated with machine learning (ML) models in determining stone-free outcomes after extracorporeal shock wave lithotripsy (ESWL), and to compare ML model performance and identify its key predictors influencing ESWL success.
Study Design: An observational study.
Place and Duration of the Study: Department of Urology, Gaziosmanpasa Training and Research Hospital, Istanbul, Turkiye, from October 2020 to November 2024.
Methodology: A total of 309 patients who underwent ESWL were analysed. The patients were categorised into stone-free and non-stone- free groups based on post-treatment imaging. Clinical parameters, including quadruple-D score (stone volume, density, skin-to-stone distance [SSD], and location), were recorded. Three ML models‒random forest (RF), logistic regression (LR), and neural network (NN)‒were trained on 80% of the dataset and tested on 20%. Model performance was assessed using accuracy, area under the curve (AUC), precision, recall, and F1 score.
Results: The quadruple-D score (AUC: 0.724) demonstrated superior predictive power compared to the Triple-D score (AUC: 0.700). Among ML models, RF achieved the highest accuracy (82.9%, AUC: 0.91), followed by NN (80.9%, AUC: 0.87) and LR (79.6%, AUC: 0.85). Significant predictors of ESWL success were stone density, volume, SSD, and the quadruple-D score, while age and body mass index (BMI) were not significant.
Conclusion: Integrating the quadruple-D score with ML models, particularly RF, enhances the prediction of ESWL outcomes. Combining clinical expertise with computational intelligence can refine patient selection and optimise treatment strategies. However, prospective studies are needed to validate these findings.

Key Words: Extracorporeal shock wave lithotripsy, Quadruple-D score, Machine learning, Random forest, Stone-free prediction.

INTRODUCTION

Extracorporeal shock wave lithotripsy (ESWL) is a widely used non-invasive treatment for kidney stones, providing a safe and effective alternative to surgery.¹ However, its success in achieving stone-free status varies due to factors such as stone composition, location, and patient-specific anatomy.² Moreover, accurate outcome prediction is essential for optimising patient selection and reducing unnecessary procedures.³

Clinical scoring systems have been developed to enhance the prediction of ESWL outcomes.⁴ The quadruple-D score integrates skin-to-stone distance (SSD), stone volume, Hounsfield unit (HU), and stone location, offering a structured assessment of stone-free likelihood.⁵

However, despite the availability of these tools, variability in ESWL outcomes underscores the need for more advanced predictive approaches.⁶

The adoption of machine learning (ML) in healthcare has revolutionised data-driven decision-making.⁷ ML algorithms can analyse large datasets and detect complex, non-linear patterns often missed by traditional methods.⁸ In urology, ML has shown promise in predicting treatment outcomes, risk stratification, and workflow optimisation.⁹ Integrating ML with established clinical scores, such as quadruple-D, could significantly enhance ESWL outcome prediction accuracy.¹⁰

Existing research has largely focused either on refining clinical scores or on applying ML separately for treatment prediction.¹¹ However, the integration of the quadruple-D score with ML remains underexplored. This convergence of clinical expertise and computational intelligence presents a promising opportunity to enhance patient care and optimise the ESWL’s clinical utility.

This study aimed to assess the effectiveness of integrating the quadruple-D score with ML algorithms in predicting stone-free rates after ESWL. By combining clinical scoring with ML, it seeks to develop a more accurate, reliable, and clinically applicable tool for optimising patient selection and treatment strategies.

METHODOLOGY

This study retrospectively analysed 350 patients who underwent ESWL at the Department of Urology, Gaziosmanpasa Training and Research Hospital, Istanbul, Turkiye, from October 2020 to November 2024. A total of 41 patients were excluded due to incomplete data, absence of pre-treatment computed tomo- graphy (CT) imaging, or other exclusion criteria, leaving 309 eligible patients. Inclusion criteria comprised patients aged 18 years or older with a solitary renal stone measuring ≤20 mm on non-contrast computed tomography (NCCT), with complete clinical and imaging data. Individuals presenting with multiple renal calculi were excluded to maintain consistency in stone- specific measurements and outcome evaluation. Exclusion criteria encompassed skeletal deformities affecting imaging accuracy, active urinary tract infections, anatomical anomalies (such as horseshoe kidney or ureteropelvic junction obstruction), and any history of urinary tract surgery within the preceding six months. Patients with positive urine cultures received appropriate antibiotic treatment and underwent ESWL only after confirmation of sterile urine. Treatment outcomes were categorised as either stone-free (defined as complete clearance with no residual fragments ≥4 mm on NCCT) or non- stone-free, indicating the presence of residual fragments ≥4 mm or the need for further intervention. This study was approved by the Ethics Committee of Gaziosmanpasa Training and Research Hospital (Decision No: 2025/07).

All patients underwent preoperative evaluation with NCCT for accurate stone assessment. NCCT was selected as it allows reliable visualisation of both radiopaque and radiolucent calculi, thereby standardising diagnostic evaluation across the study population. Key parameters, including ellipsoid stone volume (ESV), SSD, and stone density (HU), were recorded. These parameters were independently assessed by two blinded urologists. ESV was calculated using the formula π/6 × (anteroposterior diameter × transverse diameter × craniocaudal diameter), and SSD was measured according to the method described by Pareek et al.¹² The Triple-D score was calculated by assigning one point for each parameter below its respective threshold (SV <150 mm³, HU <600, SSD <12 cm), yielding scores from 0 to 3. The quadruple-D score expanded this model by incorporating stone location as an additional factor, assigning a score of 0 for lower pole stones and 1 for stones at other locations, with a total range from 0 to 4.

An ML approach was employed to evaluate the predictive capability of the quadruple-D score and clinical parameters for stone-free status. The dataset was split into training (80%) and testing (20%) subsets. Selected predictive features included demographics (age, gender, body mass index [BMI]), and preoperative imaging parameters (ESV, SSD, HU, and stone location), with the quadruple-D score included as an independent variable. Sequential session outcomes were pre-processed, with missing values imputed as 0, reflecting early treatment success. The binary outcome variable was stone-free status, defined as either complete clearance with residual fragments <4 mm (stone-free, 0) or residual fragments ≥4 mm requiring an additional intervention (non-stone-free, 1). Three ML models were developed and compared: Random forest (RF) classifier, an ensemble learning method leveraging decision trees for robust predictions; logistic regression (LR), serving as a baseline model for binary classification; and neural network (NN), a non-parametric approach capable of capturing complex data patterns. Model performance was assessed using accuracy, receiver operating characteristic (ROC)-area under the curve (AUC), precision, recall, and F1 score. The RF model was further employed for feature importance analysis, highlighting key predictors such as SSD, HU, and the quadruple-D score. Cross- validation and ROC curve analysis were performed to assess model generalisability, with statistical significance set at p <0.05 for LR.

All ESWL procedures were performed on an outpatient basis under fluoroscopic guidance to ensure precise stone localisation and optimal energy delivery. No anaesthesia was administered; instead, patients received an oral nonsteroidal anti-inflammatory drug (NSAID), Diclofenac, 30 minutes before the procedure for pain management. The Novamed Tek Novalith NT-10M lithotripter (Novamed Tek, Istanbul, Turkiye) was used. The procedure started with an initial shock frequency of 60 shocks/minute at energy level 1, gradually escalating to level 4, based on patient tolerance. The maximum shock frequency was 90 shocks/minute, with a total of 2,000–2,500 shocks per session. Higher energy levels (>4) were avoided to balance efficacy with safety, minimising the risk of renal injury while ensuring effective stone fragmentation.

Patients were monitored via kidney-ureter-bladder (KUB) radio-graphy and ultrasonography (USG) one week after each ESWL session, with a maximum of three sessions permitted. Final stone-free status was determined 12 weeks after the last session using NCCT for optimal diagnostic accuracy. Although repeated CT imaging entails radiation exposure, it was justified in this study to achieve a consistent and objective assessment of treatment efficacy across all patients. Patients were classified as stone-free if no residual fragments ≥4 mm were present; clinically insignificant residual fragments (<4 mm) were not considered treatment failures. Residual fragments ≥4 mm or the need for additional intervention classified patients as non-stone-free. NCCT was preferred for the final assessment due to its superior sensitivity, while KUB and USG were used in interim evaluations due to their practicality and lower radiation exposure.

Statistical analyses were performed using the SPSS version 27.0 (IBM Corp., Armonk, NY, USA). Categorical variables were expressed as frequencies and percentages. Continuous variables were presented as mean ± standard deviation, depending on the distribution. The normality of continuous variables was assessed using the Shapiro–Wilk test. An independent samples t-test was used to compare normally distributed variables. The chi-square or Fisher’s exact test was used for the comparison of categorical variables as appropriate. Univariate logistic regression analysis was performed to evaluate the individual association of the variables with the stone-free outcome. Variables with a p <0.05 in univariate analysis were included in multivariate logistic regression using a backward elimination approach. To prevent multicollinearity, components of the quadruple-D score (stone volume, HU, and SSD) were excluded from the final multivariate model. A p-value of <0.05 was considered statistically significant.

ML modelling and performance evaluation were conducted using Python 3.9 and the Scikit-learn 0.24 library. Three supervised ML algorithms were employed: RF, LR, and NN. RF, an ensemble learning method, constructs multiple decision trees and produces the mode of their predictions. LR was used as a baseline classifier for binary outcomes. NN was employed to capture non-linear relationships and complex data patterns. The dataset (n = 309) was randomly split into a training set (80%) and a test set (20%). Model performance was evaluated using accuracy, area under the ROC-AUC, precision, recall, and F1 score. To reduce the risk of overfitting and ensure model robustness, 5-fold cross-validation was applied during training. ROC curves were generated for model comparisons.

RESULTS

A total of 309 patients were included, of whom 192 (62.1%) were stone-free, while 117 (37.9%) had residual stones or required further intervention. There were no statistically significant differences in age (p = 0.069), BMI (p = 0.587), gender distribution (p = 0.503), preoperative UTIs (p = 0.414), or stone laterality (p = 0.555) between the groups. However, stone density, volume, and SSD were all significantly lower in the stone-free group (all p <0.001). Stone location was also significantly associated with treatment outcomes (p = 0.004). Triple-D and quadruple-D scores were significantly higher among stone-free patients (both p <0.001), whereas the non- stone-free group required slightly more ESWL sessions (p = 0.03, Table I).

Continuous variables were compared using the independent samples t-test (for normally distributed data). Categorical variables were analysed using either the Chi-square test or Fisher’s exact test, as appropriate. Bold values indicate statistical significance (p <0.05).

Table I: Patient demographics and stone characteristics.
 

Parameters	Stone-free (n = 192)	Non-stone-free (n = 117)	p-values
Age*	42.51 ± 12.89	45.17±11.71	0.069
Gender	-	-	0.503
Male	127 (66.1%)	73 (62.4%)	-
Female	65 (33.9%)	44 (37.6%)	-
BMI (kg/m2)*	27.2 ± 3.9	27.0 ± 4.7	0.587
Preoperative UTI	-	-	0.414
Yes	15 (4.9%)	13 (4.2%)	-
No	177 (57.3%)	104 (33.7%)	-
Laterality	-	-	0.555
Right	81 (26.2%)	54 (17.5%)	-
Left	111 (35.9%)	63 (20.4%)	-
Stone location	-	-	0.004
Upper calyx	20 (10.4%)	7 (6.0%)	-
Middle calyx	47 (24.5%)	18 (15.4%)	-
Lower calyx	49 (25.4%)	34 (29.1%)	-
Pelvis	37 (19.3%)	14 (12.0%)	-
Ureteropelvic junction	39 (20.3%)	44 (37.6%)	-
Stone density (HU)*	708.21 ± 233.01	895.11 ± 306.62	<0.001
Stone volume (mm3)*	406.24 ± 394.90	781.59 ± 593.64	<0.001
Stone-to-skin distance*	95.54 ± 22.67	111.06 ± 18.24	<0.001
Hydronephrosis	-	-	0.628
No	123 (39.8%)	71 (23.0%)	-
Yes	69 (22.3%)	46 (14.9%)	-
Previous treatment	-	-	0.331
Yes	26 (8.4%)	13 (4.2%)	-
No	166 (53.7%)	104 (33.7%)	-
Number of sessions*	2.3 ± 0.86	2.5 ± 0.81	0.03
Triple-D score*	1.61 ± 1.01	0.89 ± 0.73	<0.001
Quadruple-D score*	2.35 ± 0.89	1.60 ± 0.84	<0.001
*Mean ± standard deviation, BMI: Body mass index, HU: Hounsfield unit, SSD: Skin-to-stone distance. Previous treatment: Includes prior ESWL, ureteroscopy, or PCNL.

Table II: Machine learning model performance metrics.

Metrices	Random forests	Logistic regressions	Neural networks
Accuracy (%)	82.9	79.6	80.9
Precision	0.8639	0.8147	0.8276
Sensitivity (recall)	85.94%	86.98%	87.5%
Specificity	77.8%	67.5%	70.1%
F1-score	0.8616	0.8413	0.8477
ROC-AUC	0.91	0.85	0.87
AUC: Area Under the curve, F1-score: The "F" stands for "F-measure," representing the balance between precision and recall, ROC: Receiver operating characteristics.

Table III: Univariate and multivariate analysis of predictors for stone-free status.

Variables	Univariate			Multivariate
	OR	95% CI	p-value	OR	95% CI	p-values
Stone-to-skin distance	1.037	[1.024-1.050]	<0.001	-	-	-
Stone volume	1.002	[1.001-1.003]	<0.001	-	-	-
Stone density	1.003	[1.002-1.004]	<0.001	-	-	-
Triple-D score (ref ≥1.5)	5.158	[2.926-9.093]	<0.001	-	-	-
Quadruple-D score (ref ≥1.5)	5.340	[3.124-9.128]	<0.001	5.230	[3.019-9.062]	<0.001
Age	1.017	[0.999-1.037]	0.07	1.005	[0.985-1.026]	0.607
Preoperative UTI (no)	1.475	[0.675-3.221]	0.329	1.358	[0.577-3.196]	0.483
Preoperative hydronephrosis (yes)	0.866	[0.539-1.391]	0.551	0.971	[0.578-1.632]	0.913
Laterality (right)	0.851	[0.536-1.352]	0.495	-	-	-
Stone location (lower pole)	0.836	[0.500-1.399]	0.496	-	-	-
Number of sessions	1.348	[1.016-1.787]	0.03	1.319	[0.971-1.793]	0.077
OR: Odds ratio, CI: Confidence interval. p-values were calculated using binary logistic regression for univariate analysis and multivariable binary logistic regression for multivariate analysis. ORs are presented with 95% CI. Bold values indicate statistical significance (p <0.05).

Figure 1: Comparative ROC analysis of quadruple-D and triple-D scores.

Figure 2: Performance comparison: ML models vs. classic logistic regression.

The quadruple-D score exhibited superior predictive performance over the triple-D score, with significantly higher mean values in the stone-free group (p <0.001). ROC analysis was performed to determine the optimal cut-off values for the triple-D and quadruple-D scores in predicting stone-free rates. For the quadruple-D score, the optimal cut-off value was 1.50, with a sensitivity of 0.849 (84.9%) and a specificity of 0.487 (48.7%). The AUC for the quadruple-D score was 0.724, indicating a fair predictive ability. For the triple-D score, the optimal cut-off value was also 1.50, with a sensitivity of 0.500 (50.0%) and a specificity of 0.838 (83.8%). The AUC for the triple-D score was 0.700, demonstrating a moderate predictive capability (Figure 1).

ML models demonstrated strong predictive capability for stone-free status, with the RF classifier achieving the highest accuracy (82.9%) and F1 score (0.8616). Its ROC-AUC (0.91) indicated excellent discriminative power, with a sensitivity of 85.94% and a specificity of 77.8%. LR yielded an accuracy of 79.6%, an AUC of 0.85, and a higher sensitivity of 86.98% but lower specificity of 67.5%. The NN model achieved 80.9% accuracy with an AUC of 0.87, demonstrating balanced sensitivity (87.5%) and specificity (70.1%, Table II). All ML models outperform the conventional LR-based methods (Figure 2).

Multivariate LR identified the quadruple-D score as the sole independent predictor of stone-free status (OR: 5.230, 95% CI: 3.019–9.062, p <0.001). To prevent collinearity, core components of quadruple score, namely stone volume, density, and SSD, were excluded from the final model. Uni-variate analysis confirmed significant associations between stone-free status and these parameters, including both the triple-D and quadruple-D scores (all p <0.001). However, given its comprehensive integration of key predictive factors, the quadruple-D score was deemed the most clinically and statistically relevant variable. Other parameters, including age, preoperative UTI, hydronephrosis, and laterality, were not significant predictors in either univariate or multivariate analyses (Table III).

DISCUSSION

This study demonstrated the integration of the quadruple-D scoring system with ML algorithms to predict stone-free outcomes following ESWL. By combining the traditional clinical parameters of SSD, ESV, HU, and stone location with advanced computational tools, the aim was to enhance the predictive accuracy, optimise patient selection, and minimise unnecessary procedures. The findings emphasise the potential of merging clinical expertise with compu-tational intelligence to advance decision-making in urology.

Stone-related parameters are pivotal in predicting ESWL success, as underscored in the literature. Tran et al. validated the Triple-D score, which integrates SSD, ESV, and HU to optimise patient selection.⁴ While each factor individually influences ESWL outcomes, their combined assessment enhances predictive accuracy. However, the triple-D score’s linear model oversimplifies the complex interplay of these variables. Ghoneim et al. highlighted the challenges of lower pole stones, where steep infundibulo-pelvic angles and narrow infundibular diameters hinder clearance.¹³ Ichiyanagi et al. introduced the quadruple-D score by incorporating stone location, improving predictive performance (AUC: 0.651 vs. 0.596 for the Triple-D score).⁵ The current study’s findings further validate its utility, achieving an AUC of 0.724 and reinforcing the necessity of spatial parameters in ESWL outcome prediction.

The findings of this study are consistent with prior research, including that of Sengupta et al., who also validated the quadruple-D score as a superior ESWL predictor. Their study highlighted the added value of stone location in refining SFR assessments, particularly for lower pole stones, although it AUC (0.674) was slightly lower, likely due to sample size and demographic variations.¹⁴ Unlike studies identifying both SSD and BMI as key predictors, the results of this study confirmed SSD as significant, while BMI showed no statistical association. Similarly, Ozgor et al. found that SSD was predictive of ESWL success, whereas BMI lacked correlation.¹⁵ Conversely, Pareek et al. reported both as significant, with SSD having a greater impact.¹² However, these discrepancies likely stemmed from population differences and methodological variability. In this study’s cohort, a homogeneous BMI distribution may explain its statistical insignificance, underscoring the need to consider population-specific factors in ESWL predictive modelling.

Lower pole renal stones are strongly associated with poor SFR post-ESWL, as prior studies have shown. Unfavourable anatomical factors, including an obtuse infundibulo-pelvic angle, a lower calyx >1 cm, and an infundibulum <5 mm, hinder ESWL success.¹⁶ However, their routine clinical application is limited by measurement complexity and anatomical variability. Consistent with these findings, this study confirmed lower pole stones as a key determinant of ESWL success, emphasising the need to integrate stone-related and anatomical parameters into predictive models. Similarly, Ozgor et al. identified stone location and the Triple-D score as significant ESWL predictors in a multivariate analysis.¹⁵ These findings highlight the necessity of incorporating anatomical factors into clinical decision-making for improved ESWL outcome prediction.

The integration of ML into ESWL prediction marks a significant advancement in urology. Traditional regression models, constrained by linear assumptions, often fail to capture complex clinical interactions. A study has shown that ML models, particularly RF, outperform conventional methods by effectively integrating parameters such as SSD, HU, and ESV.¹⁰ In this study, RF achieved an AUC of 0.91, surpassing LR (AUC: 0.85) and NN (AUC: 0.87). RF’s ensemble learning strategy, which aggregates multiple decision trees, enhances predictive accuracy and adaptability, making it particularly suited for heterogeneous clinical datasets.¹⁷

These findings align with Eksi et al., who applied ML, including RF, to predict PICs after retrograde intrarenal surgery (RIRS), achieving an AUC of 0.956, with 87% sensitivity and 92% specificity.¹⁸ Similarly, the RF model attained an AUC of 0.91, outperforming LR. Rice et al. also demonstrated RF’s effectiveness in analysing complex clinical datasets and enhancing predictive accuracy for urinary stone treatment outcomes.¹⁰ These results underscore RF’s role in integrating computational advancements with clinical decision-making, offering both predictive power and inter-pretability through feature importance analysis, facilitating real-world application.

Park et al. explored deep learning for ESWL outcome prediction, achieving comparable accuracy but requiring larger datasets and substantial computational resources, limiting its feasibility in smaller clinical settings.¹⁹ In contrast, this study’s RF model offers an optimal balance between performance and practicality, making it well-suited for medium-sized datasets. Additionally, hybrid approaches integrating clinical expertise with ML algorithms have shown promise in optimising outcomes. Zhang et al. demonstrated that combining structured clinical data with unstructured imaging features significantly enhanced predictive perfor-mance in a similar context.²⁰

The rapid advancement of AI, including ML and deep learning, is reshaping medicine, with validated tools expected to be integrated into standard clinical workflows. Topol highlighted AI’s role in enhancing diagnostics, streamlining workflows, and enabling personalised care.²¹ In terms of improving ESWL, ML integration into hospital systems could enable real-time decision support, improving patient management, risk stratification, and complication prediction. These innovations may reduce unnecessary procedures and optimise treatment strategies. Moreover, explainable AI, as described by Ribeiro et al., could enhance clinician trust, facilitating broader adoption of ML-driven decision support tools, and advancing data-driven, personalised urology.²²

Despite its strengths, this study had some limitations. The retrospective design introduced selection bias, as only patients with complete clinical and imaging data were included, potentially skewing the results. Additionally, the exclusion of stones >20 mm and anatomical abnormalities limited the generalisability of the finding, restricting their applicability to complex cases such as staghorn calculi or congenital anomalies. These limitations underscore the need for future studies with more diverse patient cohorts to enhance model robustness.

The study’s single-centre design may have introduced biases related to institutional practices and demographic variability. While the RF model achieved an AUC of 0.91, alternative methods, such as deep learning or hybrid models, were not explored. Although deep learning can identify complex patterns, the need for large datasets and computational resources limits its feasibility in resource-constrained settings. Moreover, reliance on preoperative imaging excluded intraoperative and postoperative factors, such as patient compliance and real-time stone frag-mentation monitoring, which could further enhance predictive accuracy.

CONCLUSION

This study validated the quadruple-D score combined with ML as a robust predictor of ESWL success. Integrating clinical scoring with ML enhances accuracy and optimises patient selection and treatment planning. However, prospective multicentre studies are needed for broader validation. Incorporating real-time imaging, intraoperative variables, and patient-reported outcomes could further refine clinical applicability. As AI advances, hybrid models integrating clinical expertise with computational intelligence are expected to drive personalised medicine and streamline urological workflows.

ETHICAL  APPROVAL:
This study was approved by the Ethics Committee of Gaziosmanpasa Training and Research Hospital (Decision No: 2025/07).

PATIENTS’  CONSENT:
The authors declared that this study does not contain any personal information that could lead to the identification of patients.

COMPETING  INTEREST:
The authors declared no conflict of interest.

AUTHORS’  CONTRIBUTION:
MG, SK, MGK: Study conception, design, and manuscript preparation.
MG, AA, OC: Data collection and project administration.
SG, MGK: Analysis, supervision, and critical revision of the manuscript.
All authors approved the final version of the manuscript to be published.

REFERENCES

1. Setthawong V, Srisubat A, Potisat S, Lojanapiwat B, Pattanittum P. Extracorporeal shock wave lithotripsy versus percutaneous nephrolithotomy or retrograde intrarenal surgery for kidney stones. Cochrane Database Syst Rev 2023; 8(8):CD007044. doi: 10.1002/14651858.CD007044.pub4.
2. Shinde S, Al Balushi Y, Hossny M, Jose S, Al Busaidy S. Factors affecting the outcome of extracorporeal shockwave lithotripsy in urinary stone treatment. Oman Med J 2018; 33(3):209-17. doi: 10.5001/omj.2018.39.
3. Schmidt S, Miernik A. Extracorporeal shock wave lithotripsy versus percutaneous nephrolithotomy or retrograde intrarenal surgery for kidney stones. Urologe A 2015; 54(9):1283-6. doi: 10.1007/s00120-015-3928-7.
4. Tran TY, McGillen K, Cone EB, Pareek G. Triple D score is a reportable predictor of shockwave lithotripsy stone-free rates. J Endourol 2015; 29(2):226-30. doi: 10.1089/end. 2014.0212.
5. Ichiyanagi O, Fukuhara H, Kurokawa M, Izumi T, Suzuki H, Naito S, et al. Reinforcement of the Triple D score with simple addition of the intrarenal location for the prediction of the stone-free rate after shockwave lithotripsy for renal stones 10–20 mm in diameter. Int Urol Nephrol 2019; 51(2):239-45. doi: 10.1007/s11255-018-02066-1.
6. Yang SW, Hyon YK, Na HS, Jin L, Lee JG, Park JM, et al. Machine learning prediction of stone-free success in patients with urinary stone after treatment of shock wave lithotripsy. BMC Urol 2020; 20(1):88. doi: 10.1186/s128 94-020-00662-x.
7. Shamout F, Zhu T, Clifton DA. Machine learning for clinical outcome prediction. IEEE Rev Biomed Eng 2020; 14: 116-26. doi: 10.1109/RBME.2020.3007816.
8. Salem H, Soria D, Lund JN, Awwad A. A systematic review of the applications of expert systems and machine learning in clinical urology. BMC Med Inform Decis Mak 2021; 21(1):223. doi: 10.1186/s12911-021-01585-9.
9. Shah M, Naik N, Somani BK, Hameed BMZ. Artificial intelli-gence in urology – current use and future directions: An iTRUE study. Turk J Urol 2020; 46(Suppl 1):S27-39. doi: 10.5152/tud.2020.20117.
10. Rice P, Pugh M, Geraghty R, Hameed BZ, Shah M, Somani BK. Machine learning models for predicting stone-free status after shockwave lithotripsy: A systematic review and meta-analysis. Urology 2021; 156:16-22. doi: 10.1016/j. urology.2021.04.006.
11. Susanto AP, Lyell D, Widyantoro B, Berkovsky S, Magrabi F. Effects of machine learning-based clinical decision support systems on decision-making, care delivery, and patient outcomes: A scoping review. J Am Med Inform Assoc 2023; 30(12):2050-63. doi: 10.1093/jamia/ocad180.
12. Pareek G, Hedican SP, Lee FT Jr, Nakada SY. Shock wave lithotripsy success determined by skin-to-stone distance on computed tomography. Urology 2005; 66(5):941-4. doi: 10.1016/j.urology.2005.05.011.
13. Ghoneim IA, Ziada AM, Elkatib SE. Predictive factors of lower calyceal stone clearance after extracorporeal shockwave lithotripsy: A focus on the infundibulopelvic anatomy. Eur Urol 2005; 48(2):296-302. doi: 10.1016/j. eururo.2005.02.017.
14. Sengupta S, Basu S, Ghosh K. A prospective observational study on the predictability of Triple-D score versus Quadruple-D score in the success rate of extracorporeal shock wave lithotripsy of renal stones 1–2 cm in diameter. Urol Ann 2022; 14(1):37-42. doi: 10.4103/UA.UA_1_21.
15. Ozgor F, Tosun M, Kayali Y, Savun M, Binbay M, Tepeler A. External validation and evaluation of reliability and validity of the Triple D score to predict stone-free status after extracorporeal shockwave lithotripsy. J Endourol 2017; 31(2):169-73. doi: 10.1089/end.2016.0721.
16. Turk C, Petrik A, Sarica K, Seitz C, Skolarikos A, Straub M, et al. EAU guidelines on interventional treatment for urolithiasis. Eur Urol 2016; 69(3):475-82. doi: 10.1016/j. eururo.2015.07.041.
17. Breiman L. Random forests. Mach Learn 2001; 45:5-32. doi: 10.1023/A:1010933404324.
18. Eksi M, Fakir AE, Evren I, Sam E, Arikan Y, Kargi T, et al. Prediction of infective complications after retrograde intra renal surgery using machine learning. Minim Invasive Ther Allied Technol 2023; 32(2):73-80. doi: 10.1080/13645 706.2023.2186181.
19. Park JS, Kim DW, Lee D, Lee T, Koo KC, Han WK, et al. Development of prediction models of spontaneous ureteral stone passage through machine learning: Comparison with conventional statistical analysis. PLoS One 2021; 16(12): e0260517. doi: 10.1371/journal.pone.0260517.
20. Zhang D, Yin C, Zeng J, Yuan X, Zhang P. Combining structured and unstructured data for predictive models: A deep learning approach. BMC Med Inform Decis Mak 2020; 20(1):280. doi: 10.1186/s12911-020-01297-6.
21. Topol EJ. High-performance medicine: The convergence of human and artificial intelligence. Nat Med 2019; 25(1): 44-56. doi: 10.1038/s41591-018-0300-7.
22. Ribeiro MT, Singh S, Guestrin C. Why should I trust you? Explaining the predictions of any classifier. 22^nd ACM SIGKDD Int Conf 2016; 22:1135-44. doi: 10.1145/29396 72.2939778.