Related Experiment Video
Updated: Jun 18, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
ScLNet: A cornea with scleral lens OCT layers segmentation dataset and new multi-task model
Yang Cao1, Xiang le Yu1, Han Yao1
1Oujiang Laboratory (Zhejiang Lab for Regenerative Medicine, Vision and Brain Health), Eye Hospital and School of Ophthalmology and Optometry, Wenzhou Medical University, Wenzhou, 325000, China.
Heliyon
|July 29, 2024
Summary
This study introduces ScLNet, a deep learning model for segmenting irregular corneas and tear fluid reservoirs under scleral lenses using OCT images. ScLNet achieves high accuracy, providing a valuable tool for ophthalmic research and clinical applications.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate segmentation of corneal structures and tear fluid reservoirs (TFR) under scleral lenses is crucial for diagnosing and managing various eye conditions.
- Existing methods often struggle with irregular corneas and require manual segmentation, which is time-consuming and prone to inter-observer variability.
Purpose of the Study:
- To develop a deep learning model for accurate segmentation of irregular corneas and TFR boundaries under scleral lenses using optical coherence tomography (OCT) images.
- To create and release a publicly available dataset of cornea with scleral lens (ScL) OCT images with manual layer annotations for training and validation.
- To introduce ScLNet, a multi-task network for rapid and automated segmentation of scleral lenses, TFR, and corneal layers.
Main Methods:
- A dataset of 31,360 OCT images with scleral lens annotations was created.
- A novel multi-task deep learning network (ScLNet) was designed with an encoder featuring multi-scale input and a context coding layer, coupled with two decoders for segmentation and boundary prediction.
- Performance was evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Matthews Correlation Coefficient (MCC), Precision, and Hausdorff Distance (HD), comparing against state-of-the-art methods.
Main Results:
- ScLNet achieved high segmentation accuracy for scleral lens (98.22% DSC, 96.50% IoU), TFR (97.78% DSC, 95.66% IoU), and cornea (99.22% DSC, 98.45% IoU).
- The model demonstrated excellent boundary detection capabilities with low Hausdorff Distance (HD) values across all segmented structures.
- Layer interface recognition by ScLNet closely matched expert annotations, indicating robust performance.
Conclusions:
- The developed ScLNet model accurately segments scleral lenses, TFR, and corneal layers from OCT images, even in cases of irregular corneas.
- The publicly released ScLNet dataset and the proposed deep learning architecture offer a significant advancement for automated analysis in ophthalmic OCT imaging.
- This work facilitates improved diagnostic capabilities and research into conditions affecting the cornea and scleral lens interface.
