Stesura Seveso Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 1 ORIGINAL PAPER various outcomes, from exceedingly severe phenotypes with prenatal death to live-born individuals with accept- able renal function (1). Despite being an uncommon con- genital abnormality (prevalence 1-9/100,000), it repre- sents approximately 17% of pediatric end-stage renal dis- ease (ESRD) (1, 2). The current definitive treatment of choice is endoscopic PUV ablation; however, roughly 10% to 30% of patients require a second surgery to achieve adequate valve ablation (3, 4). Any partial outlet obstruction could cause anatomical and functional dete- rioration in the detrusor muscle of the bladder wall if it is not released quickly enough (5). This indicates the importance of close follow-up after valve ablation (5). It might be challenging to assess the result of endoscopic valve ablation. While some investigators suggest that voiding cystourethrogram (VCUG) confirms the adequacy of valve ablation, others offer cystoscopic procedures.6-8 The disadvantages of VCUG are transient dysuria, hema- turia, and toilet anxiety. On the other hand, the invasive nature and the need for admission and anesthesia are the main disadvantages of the cystoscopic procedure (4, 6). Distinguishing prognostic factors for residual valves after endoscopic valve ablation can aid in developing a better algorithm for managing PUVs and achieving purposeful follow-up for high-risk patients (4). As a result, new prediction techniques are required to allow for better patient diagnosis and follow-up (9). A study conducted by Shirazi et al. to investigate the patients with a higher risk for residual valves after PUV ablation found that younger age at surgery time, hypere- chogenicity of renal parenchyma, presence of vesi- coureteral reflux, and grade 4 or 5 reflux before surgery were associated with residual valves and need for repeat- ed ablation (4). The most commonly used form of artificial intelligence (AI) in medicine is artificial neural networks (ANNs) that mimic interconnected brain synapses and networks taught by analyzing input and output databases (10, 11). The system learns to identify variables influencing outcomes, and as more data is incorporated, self-optimization matures, resulting in more accurate predictions and higher accuracy rates for specific outcomes (12). Purpose: To build, train, and assess the arti- ficial neural network (ANN) system in esti- mating the residual valve rate after endoscopic valve ablation and compare the data obtained with conventional analysis. Methods: In a retrospective cross-sectional study between June 2010 and December 2020, 144 children with a history of poste- rior urethral valve (PUV) who underwent endoscopic valve abla- tion were enrolled in the study. MATLAB software was used to design and train the network in a feed-forward backpropagation error adjustment scheme. Preoperative and postoperative data from 101 patients (70%) (training set) were utilized to assess the impact and relative significance of the necessity for repeated ablation. The validated suitably trained ANN was used to pre- dict repeated ablation in the next 33 patients (22.9%) (test set) whose preoperative data were serially input into the system. To assess system accuracy in forecasting the requirement for repeat ablation, projected values were compared to actual outcomes. The likelihood of predicting the residual valve was calculated using a three-layered backpropagating deep ANN using preop- erative and postoperative information. Results: Of 144 operated cases, 33 (22.9%) had residual valves and needs to repeated ablation. The ANN accuracy, sensitivity, and specificity for predicting the residual valve were 90.75%, 92.73%, and 73.19%, respectively. Younger age at surgery, hyperechogenicity of the renal parenchyma, presence of vesi- coureteral reflux (VUR), and grade of reflux before surgery were among the most significant characteristics that affected postoperative outcome variables, the need for repeated ablation, and were given the highest relative weight by the ANN system. Conclusions: The ANN is an integrated data-gathering tool for analyzing and finding relationships among variables as a com- plex non-linear statistical model. The results indicate that ANN is a valuable tool for outcome prediction of the residual valve after endoscopic valve ablation in patients with PUV. KEY WORDS: Artificial neural network; Artificial intelligence; Children; Outcome; Posterior urethral Valve; Residual valve. Submitted 1 April 2024; Accepted 1 June 2024 INTRODUCTION Posterior urethral valve (PUV) is the most common cause of congenital bladder outlet obstruction in males, with Utilizing artificial neural network system to predict the residual valve after endoscopic posterior urethral valve ablation Mehdi Shirazi 1, 2, Zahra Jahanabadi 1, Faisal Ahmed 3, Davood Goodarzi 1, Alimohammad Keshtvarz Hesam Abadi 4, Mohammad Reza Askarpour 1, Sania Shirazi 5 1 Department of Urology, School of Medicine, Shiraz University of Medical Sciences, Shiraz, Iran; 2 Histomorphometry and Stereology Research Center, Shiraz University of Medical Sciences, Shiraz, Iran; 3 Department of Urology, School of Medicine, Ibb University, Ibb, Yemen; 4 PhD Candidate in Biostatistics, Department of Biostatistics, Shiraz University of Medical Sciences, Shiraz, Iran; 5 Student Research Committee, Shiraz University of Medical Sciences, Shiraz, Iran. DOI: 10.4081/aiua.2024.12530 Summary Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 M. Shirazi, Z. Jahanabadi, F. Ahmed, et al. 2 In the present study, we aimed to investigate the efficacy of ANNs as an intelligible interface to predict the rate of the residual valve after endoscopic valve ablation in children with PUV. MATERIALS AND METHODS Study design In a retrospective cross-sectional study, 144 children who were diagnosed with PUV and had undergone endoscop- ic bladder neck resection and valve ablation between June 2010 and December 2020 in the referral centers (Nemazi Teaching Hospital and Shahid Faghihi Teaching Hospital, Shiraz, southern Iran) were selected to participate in this study. The ethics committees of Shiraz University of Medical Sciences approved this project (approval code# IR.SUMS.MED.REC.1399. 585), and it was carried out in compliance with the Helsinki Declaration. All patient’s parents or legal guardians provided written informed consent before enrolment. Surgical intervention All primary PUV ablations were performed by a single experienced pediatric urologist (Prof. Shirazi). The valves were carefully ablated at the 5, 7, and 12 o'clock positions using an electrical Bugbee, and the visual assessment of the valve destruc- tion ascertained the endpoint of ablation. Then, a urethral catheter was left in place and removed 48 hours later. Patients were dis- charged with oral antibiotics. Postoperative follow-up for residual valve assessment The possible presence of residual valve rem- nants was assessed by careful clinical, radiolog- ical and endoscopic evaluation (13). For that, the patients were assessed with ultrasonography (US) every 3 to 6 months, VCUG 3 months after valve ablation, and a control cystoscopy three to 12 months later, and if a residual valve was found, a second valve ablation was per- formed (4, 14). Clinically, a persistent symp- tom of poor stream, nocturnal enuresis in older than 5 years, and persistent or worsened VUR or worsened hydronephrosis in the serial US were also considered. Slight residual valves that did not cause renal impairment or urinary tract infections (UTIs) were excluded (15). Data collection All the data used in this study, including preoperative and postoperative parameters, are summarized in Table 1. Design, training, and validation of the ANN system We designed and trained the network with MATLAB soft- ware (Mathworks, Natick, MA) using a feed-forward back- propagation error adjustment scheme (16, 17). The data were separated into training and test sets to fit the ANNs (Figure 1). Training set The network training set accounted for 101 (70%) of the study data. The architecture of the education network consists of three layers: input, hidden, and output. The number of nodes in the input and output layers corre- sponds to the number of predictor and response vari- ables. The number of hidden layer nodes was determined by trial and error. The level criterion below the charac- teristic performance curve (ROC) was used to find the best network structure. Finally, after selecting the best architecture, the network was evaluated and validated in second-stage data that did not play a role in the network training phase. Test set The network test set accounted for 43 (30%) patients. The final response variable or the final output of the model was a two-state variable with the levels of 0-no residual valve and 1-had residual valve. The number of input variables was 24, the same as the independent vari- ables. After selecting the input and output variables, a three-layer perceptron network (with an observer) was used, with one neuron in the output layer and 24 neurons in the input layer. According to the training data set and independent input variables, 320 combinations were Table 1. Variables considered for analysis. Postoperative variables Preoperative variables History of enuresis Age (Months) Blood pressure (mm Hg) Birth Weight (Low, Normal) History of urinary incontinence Time of diagnosis (Prenatal, Postnatal) Time to PUV ablation (Months) Presence of VUR (Grad) in VCUG Grade of VUR in VCUG Urinary bladder trabeculation (Mild, Moderate, Severe) in US Urinary bladder trabeculation (Mild, Moderate, Severe) in US Bladder diverticula in US Bladder diverticula in US Blood Creatinine level(mg/dL) Blood Creatinine level (mg/dL) Degree of HDN (Mild, Moderate, Severe) in US Degree of HDN (Mild, Moderate, Severe) in US Loss of cortico-medullary differentiation Loss of cortico-medullary differentiation in US Renal cortical thickness (mm) in US Renal cortical thickness (mm) in US Size of bladder wall thickness (mm) in US Size of bladder wall thickness (mm) in US History of recurrent UTI Proteinuria in urine analysis Urine culture Scar in renal DMSA scan PUV: Posterior urethral valve; DMSA: Dimercaptosuccinic acid; HDN: hydronephrosis; VUR: Vesicoureteral reflux; VCUG: voiding cystourethrogram; UTI: urinary tract infection; US: Ultrasonography. Figure 1. Schematic design of artificial neural network. Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 3 Utilizing artificial neural network system to predict the residual valve after endoscopic posterior urethral valve ablation based on 4 to 19 nodes in the hidden layer with a size of 0.8 to 0.95 (0.95, 0.9) to model a three-layer perceptron ANN. We evaluated the learning rate of 0.05 to 0.4 (0.4, 0.3, 0.2, 0.1, 0.05) with hyperbolic tangent activity in the hidden layer, sigmoid function activity in the output layer, and post-diffusion learning algorithm. After exam- ining all possible models for the structure of the three- layer ANN, the network with 24 input nodes, 10 hidden nodes, one output node, a learning rate of 0.2, and a motion size of 0.9 with an error propagation algorithm as the best ANN to predict the data. The number of fitted models for each combination was 20; the best model for each structure is shown in Table 2. Statistical analysis The mean and standard deviation were used for quantita- tive variables, and frequency and percentage were used for qualitative variables. To evaluate the predictive accu- racy of the system for each postoperative variable, the predicted values were compared with the actual out- comes (observed values), and true positive, false positive, accuracy, and precision rates of the system were calculat- ed for the need for repeated ablation. All statistical analy- ses were done using SPSS software (IBM SPSS, version 20, Armonk, New York: IBM Corp), and the significance level was also considered (p < 0.05). RESULTS The preoperative characteristics of patients are summa- rized in Table 3. The mean age at presentation was 8.65 ± 6.12 months. Low birth weight was presented in 0.3% of patients. The prenatal diagnosis of PUV was performed in 81(56.25%) patients. There were 33 (22.91%) patients who required a second valve ablation due to valve rem- nants. Postoperative data of patients are displayed in Table 4. The predictive factor for the residual valve in Regression analysis In multivariate analysis, the factors below were signifi- cantly associated with the residual valve and need for repeated ablation: younger age at operation (odds ratio [OR] 1.142; 95% Confidence interval [CI] 1.006-1.297), high initial serum creatinine (OR: 1.498; 95%CI: 1.089- 3.257), increased bladder wall thickness (OR: 1.486; 95%CI: 1.014-2.741), higher postoperative serum creati- nine (OR: 1.883; 95%CI: 1.181-4.311), presence of renal cortical thickness (OR: 1.185; 95%CI: 1.004-1.721), presence of severe bladder trabeculation (OR:1.54; 95%CI:1.136-11.281), and presence of reflux grading Table 2. Selecting the best neural network model for the recurrent data. Motion Total Area Percentage Learning Architecture Row size squares below of correct rate error ROC curve prediction 0.90 33.29 0.836 79.70 0.2 1-4-24 1 0.80 34.08 0.790 72.30 0.2 1-5-24 2 0.85 32.6 0.815 75.50 0.3 1-6-24 3 0.85 36.41 0.821 84.50 0.5 1-7-24 4 0.85 36.73 0.855 77.60 0.2 1-8-24 5 0.80 34.27 0.853 73.40 0.3 1-9-24 6 0.90 31.71 0.903 90.73 0.2 1-10-24 7 0.85 33.47 0.849 74.40 0.4 1-11-24 8 0.80 35.70 0.766 72.30 0.1 1-12-24 9 0.80 34.48 0.792 74.40 0.5 1-13-24 10 0.85 41.96 0.824 75.50 0.3 1-14-24 11 0.90 34.05 0.804 74.40 0.4 1-15-24 12 0.85 33.41 0.827 73.40 0.2 1-16-24 13 0.80 32.29 0.813 72.30 0.1 1-17-24 14 0.85 35.33 0.829 76.70 0.4 1-18-24 15 0.90 38.27 0.798 83.79 0.1 1-19-24 16 AUC: Area under the curve. Table 3. Descriptive statistics of preoperative characteristics of the patients. N (%) Variables 8.65 ± 6.12 Age at presentation (years) 5 (0.3) Low Birth Weight 81 (56.25) Prenatally Time to diagnosis 63 (43.75) Postnatally 40 (27.7) Neonatally Time of surgery 51 (35.4) 1-3 Month 14 (9.7) 3-6 Month 15 (10.4) 6-12 Month 24 (16.6) > 12 Month 69 (47.9) No Presence of VUR 15 (10.4) Left 23 (16) Right 37 (25.7) Both 8 (5.5) 1 VUR reflux grad 10 (6.95) 2 11 (7.6) 3 19 (13.2) 4 27 (18.75) 5 30 (20.8) Mild Urinary bladder trabeculation 80 (55.55) Moderate 34 (23.65) Severe 27 (18.75) Bladder diverticula 0.748 ± .618 Blood creatinine level (mg/dL) 121 (84) Prenatal KUB Sonography 31 (21.5) No Presence of hydronephrosis 11 (7.6) Right 20 (13.9) Left 82 (57) Both 31 (21.5) No Degree of hydronephrosis 18 (12.5) 1 32 (22.3) 2 35 (24.3) 3 28 (19.5) 4 42 (29.2) Loss of cortico-medullary differentiation 34 (23.6) < 3 mm Size of bladder wall thickness 87 (60.4) 3-5 mm 23 (16) > 5 mm 140 (97.2) Proteinuria in urine analysis 137 (95.1) Positive urine culture 83 (57.7) Renal scar in renal DMSA scan 26.80 ± 18.12 Left Differential function in renal DMSA scan 29.63 ± 19.75 Right AP: anterior-posterior diameter; PUV: posterior urethral valve, DMSA: dimercaptosuccinic acid; VUR: vesicoureteral reflux; UTI: urinary tract infection. Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 M. Shirazi, Z. Jahanabadi, F. Ahmed, et al. 4 three (OR: 2.526; 95%CI: 1.208-7.021), presence of reflux grading four (OR: 3.72; 95%CI:1.557-8.899). ANN model Three hundred and twenty models were fitted to the 16 structures, as shown in Table 2. We selected model 7 as the best model (green row) among these models. The pre- diction accuracy was 90.73%, and the area under the ROC curve was 0.903. The accuracy, sensitivity, and specificity of the ANN system for predicting the rate of remanent valves were 90.75%, 92.73%, and 73.19%, respectively (Table 5). The predictive factor for the residual valve in the ANN model Younger age at surgery, hyperechogenicity of the renal parenchyma, presence of VUR, and grade of reflux before surgery were among the most significant characteristics that affected postoperative outcome variables, the need for repeated ablation, and were given the highest relative weight by the ANN system (Figure 2). DISCUSSION In this study, we developed and utilized an ANN system to predict residual valve and ablation needs after endo- scopic PUV ablation. The accuracy and sensitivity ranged from 90.75% to 92.73%, proving that ANN is a valuable Table 4. Descriptive statistics of postoperative characteristics of the patients N (%) Variables 0.617 ± .429 Blood creatinine level (mg/dL) 69 (47.9) No Presence of hydronephrosis 10 (6.95) Right 18 (12.5) Left 47 (32.65) Both 69 (48) No Degree of hydronephrosis 12 (8.3) 1 21 (14.6) 2 24 (16.6) 3 18 (12.5) 4 138 (95.8) Urine analysis 136 (94.4) Urine culture 83 (57.6) Normal Blood pressure (mm Hg) 4 (2.8) Hight 17 (11.8) History of urinary incontinency 15 (10.4) History of enuresis 60 (41.6) 1 Proteinuria in urine analysis 63 (43.7) 2 21 (14.7) 3 33 (22.91) Yes Remnant valve 111 (77.09) No Figure 2. The ANN main factors that predict the residual valve. Table 5. Performance of an artificial neural network system in recurrence. AUC Accuracy Negative Positive True True predictive Predictive negative positive value value rate rate (specificity) (sensitivity) 0.903 90.75% 97.13% 50.69% 73.19% 92.73% Recurrence AUC: Area under the curve. Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 5 Utilizing artificial neural network system to predict the residual valve after endoscopic posterior urethral valve ablation tool for the prediction of the residual in PUV patients. Endoscopic primary valve ablation is the ideal initial sur- gical therapy for PUVs; however determining the absence of remaining valve remains necessitates meticulous exam- ination. Some prescribe VCUG for adequacy, while oth- ers advise cystoscopy follow-up. Both procedures offer advantages and downsides, such as transitory dysuria, enuresis, hematuria, and toileting anxiety, as well as the necessity for general anesthesia for cystoscopy. The high incidence rate of remaining leaflets necessitates post-abla- tion assessment (18). In this study, to reduce the impact of technical aspects of the first PUV ablation in the presence of remaining valves, all procedures in the research were carried out by the same surgeon with an experienced pediatric urologist with more than 20 years of experience in the pediatric urology field. Additionally, as the VCUG alone was insuf- ficient to exclude residual valve tissue (positive and neg- ative predictive values of 56% and 50%, respectively) (7), so cystoscopy was routinely performed on the patients. The reported recurrence rate of residual valves was 10%- 30% in most studies.13,19 In the same line with previous studies, the recurrence rate in this study was 22.91%. A higher recurrence rate was reported in some studies, such as the study carried out by Nawaz et al., who reported 78% residual valves during the follow-up cystoscopy. 8 The high rate of residual valves may be attributed to the use of routine follow-up cystoscopy after primary valve ablation. Previous studies have primarily examined the correlation between preoperative factors and postoperative kidney function (20, 21), with limited attention given to the risk of residual remnants valve risk in PUV patients (4). In this study, factors significantly associated with residual valve and need for repeated ablation in regression analysis include younger age at operation, high initial serum cre- atinine, increased bladder wall thickness, higher postop- erative creatinine, renal cortical thickness, severe bladder trabeculation, and advanced reflux grading (three or four). Similarly, Shirazi et al. evaluated the relationship between preoperative findings and residual obstructive leaflets after valve ablation. The authors found that younger age at PUV ablation, increased renal echogenici- ty, and the high VUR grade were significantly associated with the presence of residual valves (4). In Nabil et al.’s study, the authors found that age at presentation and sub- stantial post-void residual volume (PVR) were highly asso- ciated with the presence of residual valves (22). Motiwala et al., in their study, recommend using the urethral ratio and bladder wall thickness to assess residual valves (23). Premature birth, prenatal diagnosis, and loss of corti- comedullary differentiation unilaterally or bilaterally were all associated with meaningfully higher rates of CKD, ESRD, and the need for many corrective surgeries in PUV patients, according to Bilgutay et al. Furthermore, symp- tomatic presentation, recurrent UTI, and pre-operative and postoperative high VUR were associated with the need for many corrective surgeries but not poor renal outcomes (24). In contrast, some studies mentioned that age of presentation or ablation, urinary diversion, and the bladder neck incision did not significantly affect the out- come of PUV management (24, 25). Nevertheless, assess- ing the efficacy of the PUV ablation procedure is not uni- form and has some drawbacks (4). Additionally, bladder dysfunction after primary valve ablation is caused by the gradual deterioration of bladder contractility due to sec- ondary lower urinary tract obstruction and eventually to myogenic bladder failure (24, 25). There is a need to develop predictive models to optimize patient selection and patient counseling and to develop further ways to determine the most likely surgical out- comes before surgery is performed. In the study, younger age at surgery, hyperechogenicity of the renal parenchy- ma, presence of VUR, and grade of reflux before surgery were among the most significant characteristics that affected postoperative outcome variables, the need for repeated ablation and were given the highest relative weight by the ANN system. The advantage of ANN is that it can simply estimate and notify the relative importance of every variable on PUV outcome without the use of invasive methods. The prediction accuracy of the ANN model was 90.73%, and the area under the ROC curve was 0.903. The accuracy, sensitivity, and specificity of the system for predicting the residual valve were 90.75%, 92.73%, and 73.19%, respectively. Using a preoperative and postoperative dataset and a readily available cloud- based predictive analytical platform, we indicated that it could precisely train a model that can predict the rate of remnant leaflets valve after endoscopic valve ablation. Since predicting patient outcomes based on data sets is not a novel concept, constructing and verifying these pre- dictive models necessitates time-consuming calculations and powerful statistical techniques to be carried out by appropriately qualified professionals (26). Moreover, once these calculations were completed and the result was available and ready for presentation, the statistical data set could become out of date (27). Based on recent clinical data, the ANN system can produce these designs in a reproducible manner a few times, allowing for con- tinuous data entry, real-time constant model retraining, and provision of an output to influence clinical decision- making (28). As more input data was fed into the system (24 variables), the system innately augmented the inte- grated weights between the parameters. As a result, the predicted values were very close to the expected value. However, no studies have been conducted to date to compare the predictive accuracy of ANN versus regres- sion analysis methods in predicting which patients are at high risk of obstructive remnant leaflets after valve abla- tion in boys with PUV. Previous research on the out- comes of remnant leaflets after valve ablation reported the same predictive accuracy of ANN systems in comparison to statistical models (4, 22, 23). The advantages of the ANN model are that it is the software can be easily used and updated with more preoperative, postoperative, and intraoperative variables in future versions, and the insti- tutes can easily add other features to their software to cus- tomize it. However, the disadvantages of the ANN model include the absence of underlying causal relationships between data processing algorithms and clinical realities. Few published data investigated the role of AI in PUV outcomes. For example, Weaver et al. used ANNs to pre- dict renal failure in children with PUV using a survival analysis tool. They used deep learning imaging features Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 M. Shirazi, Z. Jahanabadi, F. Ahmed, et al. 6 from US images to predict renal failure accurately. They found that the ANNs model could accurately predict which patients will develop renal failure (29). Abdovic et al. utilized ANN to predict the late presentation of PUV in boys with lower urinary tract symptoms. The accuracy, sensitivity, specificity, positive predictive value, and neg- ative predictive value of ANNs were 92.7%, 100.0%, 89.7%, 80.0%, and 100.0%, respectively (30). Study limitation There are several limitations to the study. Firstly, the ret- rospective nature and low sample size inherently lacked control and randomization. This approach is prone to selection bias and attrition due to loss of follow-up, poten- tially skewing the results. However, such surveys can aid in the development of a better methodology for the man- agement of PUVs, as well as the achievement of purpose- ful follow-up for high-risk patients rather than invasive procedures. The assessment of outcomes was also limited, lacking adjustments for critical factors such as depth of incision, surgeon experience, urodynamic evaluation, and needs for anticholinergic therapy. Also, we do not include information on the varied outcomes, such as ESRD, dialy- sis, and renal transplantation, which can influence the outcome. This omission could lead to an incomplete understanding of the factors influencing treatment out- comes. Therefore, prospective comparative trials will be needed to confirm the predictive value of ANN vs. statis- tical data mining models and validate the system to decide whether or not to proceed with PUV ablation. CONCLUSIONS In conclusion, the ANN is an integrated data-gathering tool for analyzing and understanding the relationships among variables as a complex non-linear statistical model. The accuracy and sensitivity in predicting the rate of the resid- ual valve after the endoscopic valve ablation ranged from 90.75% to 92.73%. The results indicate that ANN might be a valuable tool for outcome prediction of the residual valve after endoscopic valve ablation in patients with PUV. ACKNOWLEDGMENTS The authors would like to thank Shiraz University of Medical Sciences, Shiraz, Iran, and also the Centre for Development of Clinical Research of Nemazee Hospital and Dr. Nasrin Shokrpour for editorial assistance. REFERENCES 1. Krishnan A, de Souza A, Konijeti R, Baskin LS. The anatomy and embryology of posterior urethral valves.J Urol. 2006; 175:1214-20. 2. Buffin-Meyer B, Tkaczyk M, Stanczyk M, et al. A single-center study to evaluate the efficacy of a fetal urine peptide signature pre- dicting postnatal renal outcome in fetuses with posterior urethral valves. Pediatr Nephrol. 2020; 35:469-75. 3. Holmdahl G, Sillen U. Boys with posterior urethral valves: out- come concerning renal function, bladder function and paternity at ages 31 to 44 years. J Urol. 2005; 174:1031-4. 4. Shirazi M, Farsiani M, Natami M, et al. Which patients are at higher risk for residual valves after posterior urethral valve ablation? Korean J Urol. 2014; 55:64-8. 5. Mirone V, Imbimbo C, Longo N, Fusco F. The detrusor muscle: an innocent victim of bladder outlet obstruction. Eur Urol. 2007; 51:57-66. 6. Bani Hani O, Prelog K, Smith GH. A method to assess posterior urethral valve ablation. J Urol. 2006; 176:303-5. 7. Smeulders N, Makin E, Desai D, et al. The predictive value of a repeat micturating cystourethrogram for remnant leaflets after pri- mary endoscopic ablation of posterior urethral valves. J Pediatr Urol. 2011; 7:203-8. 8. Nawaz G, Hussain I, Muhammad S, et al. Justification For Re- Look Cystoscopy After Posterior Urethral Valve Fulguration. J Ayub Med Coll Abbottabad. 2017; 29:30-2. 9. Lorenzo AJ, Rickard M, Braga LH, et al. Predictive Analytics and Modeling Employing Machine Learning Technology: The Next Step in Data Sharing, Analysis, and Individualized Counseling Explored With a Large, Prospective Prenatal Hydronephrosis Database.Urology. 2019; 123:204-9. 10. Rajan P, Tolley DA. Artificial neural networks in urolithiasis. Curr Opin Urol. 2005; 15:133-7. 11. Hameed BMZ, AVL SD, Raza SZ, et al. Artificial Intelligence and Its Impact on Urological Diseases and Management: A Comprehensive Review of the Literature. J Clin Med. 2021; 10 12. Aminsharifi A, Irani D, Pooyesh S, et al. Artificial Neural Network System to Predict the Postoperative Outcome of Percutaneous Nephrolithotomy. J Endourol. 2017; 31:461-7. 13. Oktar T, Salabas E, Acar O, et al. Residual valve and stricture after posterior urethral valve ablation: how to evaluate? J Pediatr Urol. 2013; 9:184-7. 14. Wu CQ, Blum ES, Patil D, Smith EA. Posterior urethral mor- phology on initial voiding cystourethrogram correlates to early renal outcomes in infants with posterior urethral valves. J Pediatr Urol. 2022; 18:813-9. 15. Gaibie Z, Mahomed N, Petersen KL, et al. Can the posterior:anterior urethral ratio on voiding cystourethrogram be used as a reliable predictor of successful posterior urethral valve ablation in male children? SA J Radiol. 2020; 24:1820. 16. Lawrence J, Luedeking S. Introduction to neural networks: design, theory and applications. Nevada City, Calif.: California Scientific Software; 1994 17. Ertin E. Mathematical Methods for Neural Network Analysis and Design: R.M. Golden, MIT Press, Cambridge, MA, 1996, 419 pp., ISBN 0-262-07174-6.J Neurocomputing. 2000; 34:257-8. 18. Mo Z, Li M, Xie X, et al. Urodynamic changes before and after endoscopic valve ablation in boys diagnosed with the posterior ure- thral valve without chronic renal failure. BMC Urol. 2023; 23:5. 19. Deshpande AV, Alsaywid BS, Smith GH. Setting the speed limit: a pilot study of the rate of serum creatinine decrease after endoscop- ic valve ablation in neonates.J Urol. 2011; 185:2497-500. 20. Klaus R, Lange-Sperandio B. Chronic Kidney Disease in Boys with Posterior Urethral Valves-Pathogenesis, Prognosis and Management.Biomedicines. 2022; 10 21. Long CJ, Bowen DK. Predicting and Modifying Risk for Development of Renal Failure in Boys with Posterior Urethral Valves. Curr Urol Rep. 2018; 19:55. 22. Nabil A, Salem A, Salah M, et al. The Importance of Second Look Cystoscopy after Posterior Urethral Valve Ablation in Children: Single Center Experience. Clin Surg. 2019; 4. Archivio Italiano di Urologia e Andrologia 2024; 96(3):12530 7 Utilizing artificial neural network system to predict the residual valve after endoscopic posterior urethral valve ablation 23. Motiwala T, Sinha A, Rathod KJ, et al. Correlation of Urethral Ratio and Bladder Wall Thickness with Cystoscopic Findings in Posterior Urethral Valve Patients to Assess Residual Valves. J Indian Assoc Pediatr Surg. 2022; 27:53-9. 24. Bilgutay AN, Roth DR, Gonzales ET, et al. Posterior urethral valves: Risk factors for progression to renal failure. J Pediatr Urol. 2016; 12:179 e1-7. 25. Hennus PM, van der Heijden GJ, Bosch JL, et al. A systematic review on renal and bladder dysfunction after endoscopic treatment of infravesical obstruction in boys. PLoS One. 2012; 7:e44663. 26. Popovics P, Penniston KL. Current research and future directions in non-malignant urologic research - proceedings of the annual CAIRIBU meeting. Am J Clin Exp Urol. 2022; 10:449-61. 27. Checcucci E, Autorino R, Cacciamani GE, et al. Artificial intelli- gence and neural networks in urology: current clinical applications. Minerva Urol Nefrol. 2020; 72:49-57. 28. Anagnostou T, Remzi M, Lykourinas M, Djavan B. Artificial neural networks for decision-making in urologic oncology. Eur Urol. 2003; 43:596-603. 29. Weaver JK, Milford K, Rickard M, et al. Deep learning imaging features derived from kidney ultrasounds predict chronic kidney dis- ease progression in children with posterior urethral valves. Pediatr Nephrol. 2023; 38:839-46. 30. Abdovic S, Cuk M, Cekada N, et al. Predicting posterior urethral obstruction in boys with lower urinary tract symptoms using deep artificial neural network.World J Urol. 2019; 37:1973-9. Correspondence Mehdi Shirazi, MD shirazim@sums.ac.ir Zahra Jahanabadi, MD (Corresponding Author) z_jahanabadi@yahoo.com Urology Office, Faghihi Hospital, Zand Blvd., Shiraz, Iran Faisal Ahmed, MD fmaaa2006@yahoo.com Davood Goodarzi, MD gdzd32@gmail.com Alimohammad Keshtvarz Hesam Abadi, PhD Candidate alimohammad.keshtvarz@gmail.com Mohammad Reza Askarpour, MD askarvip2@gmail.com Sania Shirazi, Student saniashirazi046@gmail.com Conflict of interest: The authors declare no potential conflict of interest.