Related Experiment Video
Updated: Jul 16, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
OPTIMAL TRANSPORT GUIDED UNSUPERVISED LEARNING FOR ENHANCING LOW-QUALITY RETINAL IMAGES
Wenhui Zhu1, Peijie Qiu2, Mohammad Farazi1
1School of Computing and Augmented Intelligence, Arizona State University, AZ 85281, USA.
Proceedings. IEEE International Symposium on Biomedical Imaging
|September 22, 2023
Summary
This study introduces a novel framework to enhance low-quality retinal fundus images, improving diagnostic accuracy. The method uses optimal transport and Generative Adversarial Networks (GANs) to restore image quality while preserving crucial structures.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Real-world non-mydriatic retinal fundus photography often suffers from artifacts and low quality due to patient co-morbidities.
- Image artifacts can lead to inaccurate or ambiguous clinical diagnoses, impacting patient care.
Purpose of the Study:
- To develop an effective end-to-end framework for enhancing poor-quality retinal fundus images.
- To improve the reliability and accuracy of retinal image analysis for clinical diagnosis.
Main Methods:
- Proposed a novel unpaired image-to-image translation scheme based on optimal transport theory.
- Utilized a Generative Adversarial Network (GAN) model with a generator and discriminator for image enhancement.
- Implemented an information consistency mechanism to maintain structural integrity (optic discs, blood vessels, lesions) between low-quality and enhanced images.
Main Results:
- Demonstrated the superiority of the proposed method in enhancing retinal fundus images.
- Achieved significant perceptual and quantitative improvements on the EyeQ dataset.
- Validated the effectiveness of the information consistency mechanism in preserving image structures.
Conclusions:
- The proposed framework effectively enhances poor-quality retinal fundus images, addressing a critical challenge in clinical practice.
- The method offers a promising solution for improving diagnostic accuracy through high-quality retinal imaging.
- This approach has the potential to advance automated analysis and diagnosis in ophthalmology.
