Related Experiment Video
Updated: May 25, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Geographic atrophy segmentation in infrared and autofluorescent retina images using supervised learning
K Devisetti1, T P Karnowski, L Giancardo
1University of Tennessee Health Science Center, 930 Madison Avenue, Suite 731, Memphis, TN 38163, USA.
Summary
This study developed a neural network to segment Geographic Atrophy (GA) in retinal images. The algorithm achieved high accuracy, aiding in the diagnosis of this advanced form of age-related macular degeneration (AMD).
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Geographic Atrophy (GA) is an advanced form of age-related macular degeneration (AMD).
- GA significantly contributes to legal blindness in the United States.
- Current imaging modalities like infrared (IR) and autofluorescence (AF) imaging offer complementary views of GA.
Purpose of the Study:
- To explore the use of neural network classifiers for segmenting GA.
- To evaluate the performance of these classifiers on registered IR and AF images.
Main Methods:
- Development of a neural network classifier for GA segmentation.
- Utilizing registered infrared (IR) and autofluorescence (AF) retinal images.
- Employing hold-one-out validation for performance testing.
Main Results:
- Achieved 82.5% sensitivity on a per-pixel basis.
- Achieved 92.9% specificity on a per-pixel basis.
- Demonstrated the effectiveness of the algorithm in segmenting GA.
Conclusions:
- Neural network classifiers can effectively segment Geographic Atrophy in retinal images.
- Combining IR and AF imaging enhances GA segmentation accuracy.
- The developed algorithm shows promise for diagnosing AMD-related vision loss.
