Prediction of Postnatal Fetal Renal Function: A Deep Learning Approach

Olejnik, Jeremi (2024) Prediction of Postnatal Fetal Renal Function: A Deep Learning Approach. Master's Internship Report, Biomedical Engineering.

Preview	Text mBME2024OlejnikJ.pdf Download (12MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (603kB)

Abstract

Congenital obstructive uropathies present a significant challenge in pediatric care, being the leading cause of renal failure in children. These anomalies, characterized by urinary tract obstruction (UTO), underscore the importance of early detection and intervention. Fetal renal assessment through ultrasound imaging offers the potential for timely diagnosis; however, the lack of standardized assessment methods makes it difficult to accurately predict renal function. This internship aimed to explore the integration of deep learning techniques to enhance the ultrasound-based evaluation of postnatal fetal renal function. The project was structured into three comprehensive phases: preprocessing, model development, and model evaluation. Additionally, explainable AI (XAI) techniques, specifically GradCAM, were incorporated to generate visual explanations of the model's predictions, enhancing interpretability. Despite the small size of the dataset, the results demonstrate the potential of deep learning to improve the accuracy of predicting postnatal fetal renal function from ultrasound images. The best-performing model achieved an accuracy of 68.42%, an F1 score of 66.(6)%, a sensitivity of 75%, a specificity of 63.64%, and an AUC of 69.32%. These promising outcomes highlight the need for further research and larger datasets to enhance the model's predictive capabilities before they could be used in clinical practice.

Item Type:	Thesis (Master's Internship Report)
Supervisor name:	Ooijen, P.M.A. van
Degree programme:	Biomedical Engineering
Thesis type:	Master's Internship Report
Language:	English
Date Deposited:	23 Jul 2024 12:27
Last Modified:	23 Jul 2024 12:27
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/33598

Actions (login required)

View Item