Re: Deep Learning Imaging Features Derived from Kidney Ultrasounds Predict Chronic Kidney Disease Progression in Children with Posterior Urethral Valves

Lisette A. 't Hoen*

*Corresponding author for this work

Research output: Contribution to journalComment/Letter to the editorAcademicpeer-review

Abstract

Posterior urethral valves (PUVs are a congenital anomaly that cause urinary obstruction. Owing to the obstruction, kidney development is altered, which leads to dysplasia and renal fibrosis. This can result in chronic kidney disease (CKD), with up to 20% of patients reaching end-stage kidney disease. Prenatal ultrasound has improved and prenatal detection of PUV suspicion is now more frequent, resulting in immediate renal management and early intervention after birth. However, one of the remaining clinical challenges is how to predict which children will experience progression to CKD and which will develop with normal kidney function. Previous studies have demonstrated that nadir creatinine is the best clinical factor for predicting long-term renal function. Optimal timing for determining nadir creatinine and relevant cutoff values have yet to be determined.
The current study explored whether a deep learning model can extract relevant features from the initial postnatal ultrasound that are of value in predicting progression to CKD. Deep learning models are increasingly being applied in clinical research because of their ability to detect patterns that clinicians cannot distinguish. It has been hypothesized that the pop-off mechanism due to vesicoureteral reflux may protect renal function in children with PUV. Pooling of results from relevant studies seems to suggest that this pop-off mechanism does indeed protect against progression to CKD. Might it be that the features detected via deep learning could be indicative of the presence of vesicoureteral reflux? In the current model, deep learning abilities were tested for postnatal ultrasound data. Future research models might lead us to other relevant predictive factors, such as urinary markers or prenatal investigations. The ensemble model could help in identifying which children are at greater risk of progression to CKD. In clinical practice, this could inform a more aggressive approach to prevent further kidney damage, for example, with bladder management strategies. The results should of course be seen in light of the limitation that the study is based on retrospective data. Validation of this model in different prospective cohorts is therefore necessary and might lead to new insights.
Original languageEnglish
Pages (from-to)177-178
Number of pages2
JournalEuropean Urology
Volume85
Issue number2
Early online date9 Sept 2023
DOIs
Publication statusPublished - Feb 2024

Fingerprint

Dive into the research topics of 'Re: Deep Learning Imaging Features Derived from Kidney Ultrasounds Predict Chronic Kidney Disease Progression in Children with Posterior Urethral Valves'. Together they form a unique fingerprint.

Cite this