Related Experiment Video
Updated: Jul 11, 2026
12:22
Multimodal Volumetric Retinal Imaging by Oblique Scanning Laser Ophthalmoscopy oSLO and Optical Coherence Tomography OCT
Published on: August 4, 2018
8.5K
A novel multi-scale and fine-grained network for large choroidal vessels segmentation in OCT
Wei Huang1, Qifeng Yan2, Lei Mou2
1School of Biomedical Engineering, Hainan University, Haikou, China.
Frontiers in Cell and Developmental Biology
|February 17, 2025
Summary
We developed MFGNet, a novel deep learning network for segmenting large choroidal vessels in OCT images. This method accurately identifies vessel structures, aiding in understanding choroidal diseases and high myopia.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate segmentation of large choroidal vessels in optical coherence tomography (OCT) images is crucial for quantitative analysis of choroidal diseases.
- Existing methods struggle with the small targets, low contrast, and blurred boundaries characteristic of choroidal vessels in OCT scans.
Purpose of the Study:
- To introduce MFGNet, a novel multi-scale and fine-grained network for precise segmentation of large choroidal vessels in OCT images.
- To demonstrate the clinical utility of accurate choroidal vessel segmentation in differentiating between healthy and high myopia patient groups.
Main Methods:
- Developed a two-branch fine-grained feature extraction module combining Transformer for long-range dependencies and convolution for local features, with inter-branch information exchange.
- Introduced a large kernel and multi-scale attention module to enhance features by addressing low contrast and blurred boundaries through multi-scale convolutions and feature refinement.
- Quantitatively evaluated MFGNet on 800 manually annotated OCT images, comparing its performance against state-of-the-art segmentation networks.
Main Results:
- MFGNet achieved superior performance compared to existing advanced segmentation networks on the OCT dataset.
- Three-dimensional (3D) reconstruction of large choroidal vessels was successfully performed using the segmentation outputs.
- Statistical analysis of calculated 3D morphological parameters revealed significant differences between healthy and high myopia groups.
Conclusions:
- The proposed MFGNet offers a highly effective solution for accurate large choroidal vessel segmentation in OCT images.
- The method's ability to differentiate between healthy and high myopia groups highlights its potential for clinical decision-making and disease understanding.
- This work underscores the value of advanced AI-driven image analysis in ophthalmology for quantitative disease assessment.
