Fully end-to-end automated bladder segmentation and risk assessment based on bladder morphology features for...

Area of Science:

• Urology
• Medical Imaging
• Machine Learning

Background:

• Neurogenic bladder patients often face risks of upper urinary tract dysfunction (UUTD).
• Accurate UUTD risk assessment is crucial for timely intervention.
• Current methods may lack objectivity and efficiency.

Purpose of the Study:

• To analyze bladder morphology features for UUTD risk prediction.
• To develop a fully end-to-end (E2E) automated algorithm for bladder segmentation and UUTD risk identification.

Main Methods:

• Retrospective analysis of 604 neurogenic bladder patients undergoing video-urodynamics (VUDS).
• Extraction of 36 morphology features (shape, LoG histograms, fractal characteristics) from manually and automatically segmented bladder regions of interest (ROIs).
• Development and evaluation of machine learning models (e.g., SVM) using selected features and an FCN-ResNet101 for automated segmentation.

Main Results:

• 28 morphology features significantly differed between bladder grades, with Shape_Perimeter_to_Surface_ratio being a key UUTD predictor (AUC=0.722).
• The SVM model achieved high performance (AUC=0.860 training, 0.830 testing) outperforming single features.
• Automated segmentation (Dice=0.9848, mIoU=0.9705) correlated strongly with manual segmentation (r>0.90), enabling an E2E approach with comparable ML model performance (AUC=0.843 training, 0.815 testing).

Conclusions:

• Bladder morphology features are robust predictors of UUTD.
• The developed E2E automated model offers a quantitative, objective tool for early UUTD risk assessment, outperforming traditional clinical parameters.
• Further validation in large-scale, multicenter prospective studies is recommended to confirm clinical applicability.