ARTIFICIAL NEURAL NETWORKS IN PEDIATRIC UROLOGY: PREDICTION OF SONOGRAPHIC OUTCOME FOLLOWING PYELOPLASTY

Bioinformatics in urology - molecular characterization of pathophysiology and response to treatment

The application of bioinformatics has revolutionized the practice of medicine in the past 20 years. From early studies that uncovered subtypes of cancer to broad efforts spearheaded by the Cancer Genome Atlas initiative, the use of bioinformatics strategies to analyse high-dimensional data has provided unprecedented insights into the molecular basis of disease. In addition to the identification of disease subtypes - which enables risk stratification - informatics analysis has facilitated the identification of novel risk factors and drivers of disease, biomarkers of progression and treatment response, as well as possibilities for drug repurposing or repositioning; moreover, bioinformatics has guided research towards precision and personalized medicine. Implementation of specific computational approaches such as artificial intelligence, machine learning and molecular subtyping has yet to become widespread in urology clinical practice for reasons of cost, disruption of clinical workflow and need for prospective validation of informatics approaches in independent patient cohorts. Solving these challenges might accelerate routine integration of bioinformatics into clinical settings.

Deep learning based automated quantification of urethral plate characteristics using the plate objective scoring tool (POST)

INTRODUCTION

The plate objective scoring tool (POST) was recently introduced as a reproducible and precise approach to quantifying urethral plate (UP) characteristics and guide to selecting particular surgical techniques. However, defining the landmarks mandatory for the POST score from captured images can potentially leads to variability. Although artificial intelligence (AI) is yet to be wholly accepted and explored in hypospadiology, it has certainly brought new possibilities to light.

OBJECTIVES

To explore the capacity of deep learning algorithm to further streamline and optimize UP characteristics appraisal on 2D images using the POST, aiming to increase the objectivity and reproducibility of UP appraisal in hypospadias repair.

METHODS

The five key POST landmarks were marked by specialists in a 691-image dataset of prepubertal boys undergoing primary hypospadias repair. This dataset was then used to develop and validate a deep learning-based landmark detection model. The proposed framework begins with glans localization and detection, where the input image is cropped using the predicted bounding box. Next, a deep convolutional neural network (CNN) architecture is used to predict the coordinates of the five POST landmarks. These predicted landmarks are then used to assess UP characteristics in distal hypospadias.

RESULTS

The proposed model accurately localized the glans area, with a mean average precision (mAP) of 99.5% and an overall sensitivity of 99.1%. A normalized mean error (NME) of 0.07152 was achieved in predicting the coordinates of the landmarks, with a mean squared error (MSE) of 0.001 and a 2.5% failure rate at a threshold of 0.2 NME.

DISCUSSION

Our results support the possibility of further standardizing UP assessment from captured hypospadias images, and the use of machine learning algorithms and image recognition shows that these novel artificial intelligence technologies are useful for scoring hypospadias. External validation can provide valuable information on the generalizability and reliability of deep learning algorithms, which can aid in assessments, decision-making and predictions for surgical outcomes.

CONCLUSIONS

This deep learning application shows robustness and high precision in using POST to appraise UP characteristics. Further assessment using international multi-centre image-based databases is ongoing.

Artificial neural networks for assessing the risk of urinary calcium stone among men

The pathophysiology of idiopathic calcium oxalate nephrolithiasis involves metabolic abnormalities. Previous studies gave conflicting results about the metabolic factors in stone formers. Artificial neural networks (ANN) are new methods based on computer programming that have outperformed conventional methods in prediction of outcomes in different medical applications. The aim of our study was to compare with ANN the clinical and biochemical parameters implicated in urinary calcium stone between stone formers (SF) and controls (C). We compared 11 clinical and biochemical variables among 119 male idiopathic calcium oxalate SF and 96 C by univariate and multivariate statistical analyses. Univariate analyses included discriminant analysis, logistic regression analysis, and ANN. For multivariate analyses, stepwise discriminant analysis and ANN were performed. Variables included age, body mass index (BMI), family history of nephrolithiasis, supersaturation with respect to calcium oxalate, calcemia, and 24-h urinary calcium, oxalate, uric acid, urea, sodium, and citrate. With univariate and multivariate analyses, ANN were as efficient as classical statistical analyses in discriminating the different parameters. The sensitivity, the specificity, and the percentage of correctly classified patients were similar in all analyses. With ANN, supersaturation (receiver operating characteristic, ROC curves index 0.73) and urea (ROC 0.72) were the most discriminant followed by family history and urinary calcium (ROC 0.67). ROC index was 0.63 for citrate, 0.61 for oxalate and urate, 0.60 for sodium and calcemia, 0.58 for age, and 0.56 for BMI, but these parameters were not statistically different between SF and C. ANN gave additional information since they made it possible to determine the cut-off values of the parameters and their predictive power. Cut-off values for being a stone former were 8.9 for supersaturation and 363 mmol/day for urinary urea with a predictive power of 0.69 and 0.70, respectively. Univariate and multivariate analysis evidenced supersaturation and 24-h urinary urea excretion as the most discriminant parameters between the two populations. In addition to high supersaturation, the negative impact of protein intake was confirmed.

A novel approach for accurate prediction of spontaneous passage of ureteral stones: Support vector machines

The objective of this study was to optimally predict the spontaneous passage of ureteral stones in patients with renal colic by applying for the first time support vector machines (SVM), an instance of kernel methods, for classification. After reviewing the results found in the literature, we compared the performances obtained with logistic regression (LR) and accurately trained artificial neural networks (ANN) to those obtained with SVM, that is, the standard SVM, and the linear programming SVM (LP-SVM); the latter techniques show an improved performance. Moreover, we rank the prediction factors according to their importance using Fisher scores and the LP-SVM feature weights. A data set of 1163 patients affected by renal colic has been analyzed and restricted to single out a statistically coherent subset of 402 patients. Nine clinical factors are used as inputs for the classification algorithms, to predict one binary output. The algorithms are cross-validated by training and testing on randomly selected train- and test-set partitions of the data and reporting the average performance on the test sets. The SVM-based approaches obtained a sensitivity of 84.5% and a specificity of 86.9%. The feature ranking based on LP-SVM gives the highest importance to stone size, stone position and symptom duration before check-up. We propose a statistically correct way of employing LR, ANN and SVM for the prediction of spontaneous passage of ureteral stones in patients with renal colic. SVM outperformed ANN, as well as LR. This study will soon be translated into a practical software toolbox for actual clinical usage.

Multi-institutional validation study of neural networks to predict duration of stay after laparoscopic radical/simple or partial nephrectomy

PURPOSE

We developed models to predict post-laparoscopic radical or simple nephrectomy (LapNx) and post-laparoscopic partial nephrectomy (LapPNx) hospital duration of stay (DOS).

MATERIALS AND METHODS

We performed a retrospective review (design group) of all 726 patients (July 1997 to April 2004) who underwent LapNx or LapPNx at the Cleveland Clinic Foundation (CCF). Preoperative findings were recorded. Neural network algorithms were designed to predict the DOS before surgery. The models were then tested on a separate 252 patients from 6 different institutions, namely Tulane University Medical School, University of Arkansas for Medical Sciences, Cedars-Sinai Medical Center, University of Iowa, Mayo Clinic at Scottsdale and CCF.

RESULTS

In the CCF design groups, the LapNx model accuracy was 73% to 74% and the LapPNx model 73% to 83%. Overall accuracy in the test groups at all 6 institutions was 72% (area under ROC 0.6 to 0.7) for the LapNx model and 52% to 81% (ROC 0.5 to 0.7) for the LapPNx model.

CONCLUSIONS

The LapNx model provides 72% accuracy in predicting the DOS at all 6 institutions. The LapPNx model provided fair accuracy only at CCF and Tulane University Medical School. These models may streamline the delivery of care and continued testing will allow for further refinement.