Related Experiment Video
Updated: Jun 22, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Automated eyeball volume measurement based on CT images using neural network-based segmentation and simple estimation
Sujeong Han1, Jeong Kyu Lee2, Daewon Lee3
1Department of Artificial Intelligence, Chung-Ang University, Seoul, 06974, Republic of Korea.
Scientific Reports
|July 2, 2024
Summary
A new automated method accurately estimates eyeball volume from CT scans, incorporating eyeball shape for better myopia progression monitoring. This technique improves upon existing methods for diagnosing ocular diseases.
Area of Science:
- Ophthalmology
- Medical Imaging
- Biomedical Engineering
Background:
- Increasing digital device use correlates with rising myopia incidence, a risk factor for ocular diseases.
- Eyeball volume is linked to myopia, making its accurate estimation crucial for monitoring progression.
- Current eyeball volume estimation methods often neglect the actual eyeball's complex shape, limiting accuracy.
Purpose of the Study:
- To develop an automated method for estimating eyeball volume from computed tomography (CT) images.
- To integrate prior knowledge of eyeball shape into the volume estimation process.
- To improve the accuracy of eyeball volume estimation for monitoring myopia progression and aiding ocular disease diagnosis.
Main Methods:
- A multi-step process involving data preprocessing, image segmentation, and volume estimation.
- Utilized various deep learning models (U-Net, HFCN, DeepLab v3+, SegNet, HardNet-MSEG) for eyeball image segmentation.
- Employed a combination of the truncated cone formula and integral equation for volume calculation, validated against manual estimations from 200 subjects.
Main Results:
- U-Net demonstrated superior performance among the tested segmentation models.
- The proposed automated eyeball volume estimation method significantly outperformed comparative techniques.
- Achieved a high correlation coefficient (0.819), low mean absolute error (0.640), and low mean squared error (0.554).
Conclusions:
- The developed method provides accurate eyeball volume estimation by accounting for eyeball shape.
- It offers a valuable tool for monitoring myopia progression and potentially aids in diagnosing various ocular diseases.
- The methodology shows promise for extension to volume estimation of other ocular structures.
