BlaVeS: A Novel Hand-Labeled Dataset for Improved Bladder Vessel Segmentation with Modified U-Net.

Area of Science:

• Urology
• Medical Imaging
• Computer Vision

Background:

• Accurate localization of urinary bladder abnormalities is crucial for uro-oncology treatments.
• Blood vessels are key anatomical landmarks for orientation during cystoscopy, but segmentation is challenging due to variable conditions.
• Existing research on bladder vessel segmentation is limited.

Purpose of the Study:

• To introduce the first publicly available dataset for bladder vessel segmentation from endoscopic images.
• To develop and evaluate a deep learning model for improved bladder vessel segmentation.
• To address limitations of standard segmentation architectures in the context of uro-oncology.

Main Methods:

• Creation of a manually annotated dataset of endoscopic bladder images, including various cystectomy scenarios.
• Adaptation of a U-Net architecture with deep skip connections and attention mechanisms.
• Evaluation of the modified U-Net model on the new dataset.

Main Results:

• Standard retinal vessel segmentation models performed poorly on the bladder dataset.
• The proposed modified U-Net achieved an accuracy of 0.93 and a precision of 0.55.
• The developed method demonstrates improved performance under challenging imaging conditions.

Conclusions:

• The introduced dataset and modified U-Net architecture advance automated bladder vessel segmentation.
• This work provides a foundation for improved diagnostic and therapeutic outcomes in urological imaging.
• The findings facilitate more precise patient-specific treatments in uro-oncology.