BWS-Net: An Optimal Deep Learning Architecture for the Anterior Bladder Wall Segmentation using Ultrasound Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Urodynamic tests assess bladder function via invasive detrusor pressure measurement.
• Ultrasound bladder vibrometry offers non-invasive bladder compliance evaluation.
• Accurate anterior bladder wall segmentation is crucial for these non-invasive methods.

Purpose of the Study:

• To develop a novel deep learning network for precise anterior bladder wall segmentation.
• To improve upon existing segmentation methods in terms of accuracy and efficiency.
• To enable real-time clinical applications for bladder function assessment.

Main Methods:

• Proposed a deep learning network with blueprint separable convolutions and adaptive attention-based skip connections.
• Employed an encoder-decoder structure for segmentation.
• Evaluated performance using 5-fold cross-validation on 8592 images from 64 patients.

Main Results:

• Achieved a mean Dice score of 0.82 and sensitivity of 0.85.
• Reported a mean root mean square error of 0.67 ± 0.35mm for bladder wall thickness.
• Demonstrated improved segmentation performance and significant reductions in computational complexity (94-96%) compared to existing methods.

Conclusions:

• The proposed deep learning network enables accurate anterior bladder wall segmentation.
• Blueprint separable convolutions and adaptive attention-skip connections enhance performance and reduce computational load.
• The method shows potential for effective, real-time clinical application in bladder function assessment.