Related Experiment Video
Updated: Aug 17, 2025
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
21.7K
An efficient registration-based approach for retinal blood vessel segmentation using generalized Pareto and fatigue
K Susheel Kumar1, Nagendra Pratap Singh2
1GITAM University, Bengaluru, 561203, India; National Institute of Technology Hamirpur, Himachal Pradesh 177005, India.
Medical Engineering & Physics
|December 18, 2022
Summary
This study introduces a novel method for segmenting and registering retinal blood vessels using Generalized Pareto probability distribution functions and Binary Robust Invariant Scalable Key points. The approach achieves high accuracy in detecting changes for diagnosing eye diseases.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Accurate segmentation and registration of retinal blood vessels (RBV) are crucial for diagnosing conditions like glaucoma, diabetes, and hypertension.
- Identifying changes in RBV structure is challenging due to low-intensity backgrounds in retinal images and limitations of existing methods.
- Past methods have struggled to identify specific network failures within the RBV structure.
Purpose of the Study:
- To develop a novel approach for segmenting and registering retinal blood vessels for improved diagnostic capabilities.
- To accurately identify changes in retinal blood vessel structure for early disease detection.
- To overcome the challenges posed by low-intensity backgrounds in retinal image analysis.
Main Methods:
- A novel segmentation approach utilizing a matched filter of the Generalized Pareto Probability Distribution Function (pdf).
- A feature-based registration approach employing Binary Robust Invariant Scalable Key points (BRISK) for attention point recognition and matching.
- A three-level process involving pre-processing, matched filter-based pdf segmentation, and BRISK registration.
Main Results:
- The proposed Generalized Pareto pdf approach achieved an average accuracy of 95.21% for segmentation.
- Normalized Cross-Correlation (NCC) between image pairs was 93%, indicating robust segmentation.
- The BRISK-based registration achieved an average accuracy of 98.51% for identifying changes in vessel structure.
- The system demonstrated an average computational time of approximately 1.4 seconds.
Conclusions:
- The novel approach effectively segments and registers retinal blood vessels with high accuracy.
- This method offers a significant improvement for the early detection and diagnosis of various eye diseases.
- The combination of Generalized Pareto pdf and BRISK features provides a powerful tool for retinal image analysis.
