Related Experiment Video
Updated: Jul 2, 2025
06:15
Author Spotlight: Anterior HR-OCT as a Non-Invasive Tool for Characterizing Ocular Surface Squamous Neoplasia
Published on: August 9, 2024
1.2K
SeqCorr-EUNet: A sequence correction dual-flow network for segmentation and quantification of anterior segment OCT
Jing Fang1, Aoyu Xing1, Ying Chen2
1School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, 230601, Anhui, China.
Computers in Biology and Medicine
|February 16, 2024
Summary
A novel deep learning model, SeqCorr-EUNet, accurately segments anterior segment optical coherence tomography (AS-OCT) images. This improves analysis of ophthalmic diseases and quantification of clinical parameters, even in low-quality images.
Area of Science:
- Ophthalmic imaging analysis
- Medical image segmentation
- Artificial intelligence in ophthalmology
Background:
- Accurate segmentation of anterior segment optical coherence tomography (AS-OCT) images is crucial for analyzing ophthalmic disease morphology and extracting clinical parameters.
- Manual segmentation is labor-intensive, prone to errors, and challenged by image noise and artifacts.
- Existing methods struggle with low-quality AS-OCT images, hindering precise clinical assessments.
Purpose of the Study:
- To introduce SeqCorr-EUNet, a novel dual-flow deep learning model for semantic segmentation and quantification of AS-OCT images.
- To enhance the extraction of intra-slice and inter-slice features for improved segmentation accuracy.
- To enable automatic extraction of clinical parameters, 3D reconstruction, and volume measurement of the anterior segment.
Main Methods:
- Developed a dual-flow architecture incorporating an EfficientNet encoder for intra-slice feature extraction.
- Utilized a convolutional gated recurrent unit (ConvGRU) based sequence correction flow for inter-slice feature extraction.
- Integrated a channel attention gate in skip-connections to refine contextual information and reduce noise.
- Fused spatio-temporal information to correct pre-segmentation morphological details.
Main Results:
- SeqCorr-EUNet demonstrated competitive performance compared to existing methods on a public AS-OCT dataset.
- The model significantly improved segmentation and quantification accuracy, particularly for low-quality AS-OCT images.
- Achieved automatic extraction of clinical parameters, 3D anterior chamber reconstruction, and volume measurement.
Conclusions:
- SeqCorr-EUNet offers a robust and accurate solution for AS-OCT image segmentation and quantification.
- The proposed model enhances diagnostic capabilities for ophthalmic diseases by improving morphological analysis and parameter extraction.
- This approach shows significant potential for clinical application in ophthalmology, especially in challenging imaging scenarios.
