Identifying urethral strictures using machine learning: a proof-of-concept evaluation of convolutional neural network...

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Retrograde urethrograms (RUG) are essential for evaluating urethral strictures and planning reconstructive surgery.
• Current methods rely on expert interpretation of RUG images.
• Developing automated tools can enhance diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN)-based machine learning algorithm.
• To accurately characterize RUG images, distinguishing between urethral strictures and normal anatomy.
• To assess the feasibility of using AI for RUG image analysis.

Main Methods:

• Collected and anonymized RUG images from multiple institutions and online sources.
• Developed a CNN model with three convolutional and three pooling layers.
• Employed data augmentation techniques and split data into 90% training and 10% testing sets.
• Trained the model for one hundred epochs.

Main Results:

• The study analyzed 242 RUG images (196 strictures, 46 normal).
• The CNN model achieved 92.2% accuracy on the training dataset.
• Validation accuracy on the testing set was 88.5% for correct image characterization.

Conclusions:

• A machine learning algorithm can effectively differentiate between normal and stricture RUG images.
• Further model development could enable characterization of stricture location and length.
• This AI approach holds potential for optimizing surgical repair strategies for urethral strictures.