Related Experiment Video
Updated: Oct 22, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
PixelBNN: Augmenting the PixelCNN with Batch Normalization and the Presentation of a Fast Architecture for Retinal
Henry A Leopold1, Jeff Orchard2, John S Zelek1
1Department of Systems Design Engineering, University of Waterloo, Waterloo, ON N2L 3G1, Canada.
Journal of Imaging
|August 30, 2021
Summary
This study introduces PixelBNN, a faster deep learning method for segmenting retinal vessels in fundus images. It achieves comparable performance to state-of-the-art methods with significantly reduced computation time.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Retinal fundus image analysis is crucial for diagnosing and treating eye conditions.
- Accurate segmentation of retinal vasculature aids in disease assessment and surgical planning.
- Automated segmentation faces challenges due to image noise and variations.
Purpose of the Study:
- To evaluate key performance indicators (KPIs) and current methods for retinal vessel segmentation.
- To analyze computational efficiency-performance trade-offs with information loss.
- To introduce PixelBNN, an efficient deep learning model for automated fundus morphology segmentation.
Main Methods:
- PixelBNN model development and training.
- Testing and cross-testing on DRIVE, STARE, and CHASE_DB1 datasets.
- Performance evaluation using G-mean, Mathews Correlation Coefficient, and F1-score.
Main Results:
- PixelBNN demonstrated 8.5x faster test-time computation than state-of-the-art methods.
- Achieved comparable performance despite a 5x to 19x reduction in image information.
- Efficiency gains were observed with varying degrees of information loss.
Conclusions:
- PixelBNN offers a highly efficient solution for automated retinal vessel segmentation.
- The model provides a favorable balance between computational speed and segmentation accuracy.
- This approach has potential for improving diagnostic workflows in ophthalmology.
