Related Experiment Video
Updated: May 26, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
348
Wide-field OCT volumetric segmentation using semi-supervised CNN and transformer integration.
Syna Sreng1,2, Padmini Ramesh1,2, Pham Duc Nam Phuong3
1Singapore Eye Research Institute, Singapore National Eye Centre, Singapore City, Singapore.
Scientific Reports
|February 24, 2025
Summary
This study introduces a novel semi-supervised learning framework for wide-field optical coherence tomography (OCT) retinal layer segmentation. The method significantly improves accuracy by efficiently using labeled and unlabeled data, outperforming existing models.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Wide-field optical coherence tomography (OCT) enables peripheral retinal monitoring but poses challenges for accurate layer segmentation.
- Deep Convolutional Neural Networks (CNNs) require extensive annotated data, limiting their application in complex medical imaging tasks.
Purpose of the Study:
- To develop and evaluate an advanced semi-supervised learning framework for robust wide-field retinal layer segmentation.
- To reduce the dependency on large, manually annotated datasets for OCT image analysis.
Main Methods:
- A semi-supervised learning framework combining Convolutional Neural Networks (CNNs) and transformers was proposed.
- The model was trained and evaluated on a dataset of 74 wide-field OCT scans (15x9 mm FOV), including 11,750 labeled and 29,016 unlabeled images.
- Performance was compared against a UNet baseline and validated using a clinical spectral-domain-OCT system.
Main Results:
- The proposed semi-supervised method demonstrated significant improvements in wide-field retinal layer segmentation, achieving up to 11% enhancement over the UNet baseline (P < 0.001).
- Retinal layer thickness measurements showed high correlation (up to 0.91) with a clinical spectral-domain-OCT system (P < 0.001).
Conclusions:
- Semi-supervised learning, leveraging cross-teaching between CNNs and transformers, is highly effective for automated wide-field OCT layer segmentation.
- This approach enhances the monitoring of peripheral retinal changes and improves the accuracy of retinal layer thickness measurements.
