Related Experiment Video
Updated: Oct 2, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Application of Artificial Intelligence and Deep Learning for Choroid Segmentation in Myopia
Hung-Ju Chen1, Yu-Len Huang2, Siu-Lun Tse2
1Department of Ophthalmology, Taichung Veterans General Hospital, Taichung, Taiwan.
Translational Vision Science & Technology
|February 25, 2022
Summary
Deep learning accurately segments the choroid, revealing it is thinner in high myopia. This method aids in evaluating chorioretinal diseases and myopia progression.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Myopia progression is a growing concern, and understanding its relationship with ocular structures like the choroid is crucial.
- Accurate measurement of choroidal thickness is essential for myopia research and clinical evaluation.
- Automated methods for analyzing optical coherence tomography (OCT) images can improve efficiency and consistency.
Purpose of the Study:
- To develop and validate a deep learning model for automated choroidal segmentation and thickness measurement from OCT images.
- To investigate the correlation between choroidal thickness and myopia, particularly in high myopia cases.
- To assess the clinical utility of the developed model in quantifying choroidal changes.
Main Methods:
- A mask region-based convolutional neural network (R-CNN) model was trained using manually labeled OCT images.
- The model was trained with pretrained weights and validated on a separate dataset.
- Choroidal thickness was automatically segmented and quantified, and its relationship with myopia was analyzed.
Main Results:
- The mask R-CNN model demonstrated high accuracy in choroidal segmentation, with low boundary segmentation errors (6.72 ± 2.12 µm inner, 13.75 ± 7.57 µm outer).
- The mean Dice coefficient between automatic and manual segmentation was 93.87% ± 2.89%, indicating excellent agreement.
- Choroidal thickness was found to be significantly thinner in high-myopic eyes, with axial length being the most significant predictor.
Conclusions:
- The mask R-CNN model is highly effective for choroidal segmentation and quantification in OCT images.
- High myopia is associated with significantly thinner choroids compared to non-high myopia.
- This deep learning approach provides a foundation for evaluating complex chorioretinal diseases and monitoring myopia progression.
