Related Experiment Video
Updated: Sep 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
A Deep Learning Model for Automated Segmentation of Geographic Atrophy Imaged Using Swept-Source OCT
Varsha Pramil1, Luis de Sisternes2, Lars Omlor2
1Tufts University School of Medicine, Boston, Massachusetts; New England Eye Center, Tufts New England Medical Center, Boston, Massachusetts.
Ophthalmology. Retina
|August 15, 2022
Summary
A new deep learning algorithm accurately segments geographic atrophy (GA) using OCT images, providing reproducible measurements for assessing GA growth over time. This automated method matches manual grading accuracy for key GA metrics.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Geographic atrophy (GA) is a leading cause of vision loss.
- Accurate measurement of GA progression is crucial for clinical trials and patient management.
- Current methods for GA assessment can be time-consuming and subjective.
Purpose of the Study:
- To develop and validate a deep learning algorithm for automated segmentation of geographic atrophy (GA) using en face swept-source optical coherence tomography (SS-OCT) images.
- To assess the accuracy and reproducibility of the algorithm in measuring GA area and enlargement rate.
Main Methods:
- A deep learning algorithm was developed using SS-OCT scan volume data, incorporating features like hypertransmission, RPE loss, and retinal thickness loss.
- The algorithm was trained on 126 images and tested on 180 scans from 30 eyes with GA and 45 images from 42 eyes without GA.
- Measurements of GA area and enlargement rate were compared to manual grading, with repeatability assessed using intraclass coefficients (ICCs).
Main Results:
- The automated algorithm demonstrated no significant differences compared to manual graders for GA area and enlargement rate measurements.
- High repeatability was observed, with ICCs of 0.99 for GA area and 0.94 for enlargement rate.
- The algorithm achieved high accuracy in delineating GA and measuring its progression.
Conclusions:
- The proposed deep learning-based automated algorithm provides accurate and reproducible segmentation of GA from en face SS-OCT images.
- This tool facilitates reliable assessment of GA growth over time, aiding in clinical research and patient care.
- The algorithm offers a promising automated solution for objective GA quantification.
