Related Experiment Video
Updated: Dec 20, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Fully Automated Segmentation of Globes for Volume Quantification in CT Images of Orbits using Deep Learning
L Umapathy1,2, B Winegar2, L MacKinnon2
1From the Departments of Electrical and Computer Engineering (L.U., A.B.).
AJNR. American Journal of Neuroradiology
|May 23, 2020
Summary
This study introduces a deep learning model for fast and accurate quantification of ocular globe volumes from CT scans, aiding in trauma diagnosis. The automated method shows high accuracy comparable to human experts.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Ophthalmology
Background:
- Accurate quantification of globe volumes is crucial for diagnosing ocular trauma.
- Current methods may be time-consuming or lack automation.
- Computed tomography (CT) is a key imaging modality for orbital evaluation.
Purpose of the Study:
- To develop and validate an automated deep learning workflow for predicting globe contours and quantifying volumes in CT images.
- To assess the accuracy and reliability of the proposed deep learning model against manual segmentation.
Main Methods:
- A 2D Modified Residual UNET (MRes-UNET2D) deep learning model was developed.
- The model was trained on axial CT images of 80 subjects without globe injuries.
- Performance was evaluated using Dice, precision, recall, and Hausdorff distance on two independent test cohorts.
Main Results:
- The MRes-UNET2D model achieved high accuracy with an average Dice score of 0.95.
- Volume estimation error was low (5.3%), with no significant difference from ground truth.
- Model performance approached human inter-observer variability in contour delineation.
Conclusions:
- The automated deep learning workflow provides fast and reliable globe volume quantification from CT images.
- This method can aid in the diagnosis and management of ocular trauma.
- The approach is robust across various CT acquisition parameters.
