Related Experiment Video
Updated: Dec 18, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Automated Retinal Layer Segmentation Using Graph-based Algorithm Incorporating Deep-learning-derived Information
Zubin Mishra1, Anushika Ganegoda1, Jane Selicha1
1Doheny Image Analysis Laboratory, Doheny Eye Institute, Los Angeles, CA, 90033, USA.
Scientific Reports
|June 14, 2020
Summary
A new algorithm automatically segments retinal layers and early age-related macular degeneration (AMD) features like reticular pseudodrusen (RPD) and regular drusen in OCT scans, improving diagnostic accuracy.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Regular drusen are established early indicators of nonneovascular age-related macular degeneration (AMD).
- Reticular pseudodrusen (RPD), also known as subretinal drusenoid deposits (SDD), are increasingly recognized extracellular deposits above the retinal pigment epithelium (RPE) in AMD.
- Accurate segmentation of retinal layers and these distinct drusen types is crucial for AMD diagnosis and monitoring.
Purpose of the Study:
- To develop and validate an automated algorithm for segmenting retinal layers and differentiating between regular drusen and RPD in spectral domain optical coherence tomography (SD-OCT) images.
- To establish a robust computational method for analyzing early AMD features in high-resolution OCT data.
Main Methods:
- A shortest-path algorithm, enhanced by probability maps from a fully convolutional neural network, was employed for segmentation.
- The algorithm was designed to segment 11 distinct retinal layers along with both regular drusen and RPD in SD-OCT volumes.
- Validation focused on achieving subpixel accuracy for the segmented structures.
Main Results:
- The developed algorithm successfully segmented regular drusen, RPD, and 11 retinal layers with high accuracy.
- The mean difference in segmentation was within the subpixel accuracy range, demonstrating algorithm robustness.
- This represents the first validated algorithm for the separate, automated segmentation of retinal layers and early AMD features (RPD and regular drusen) on SD-OCT.
Conclusions:
- Automated segmentation of retinal layers and early AMD features like RPD and regular drusen is feasible using advanced deep learning and shortest-path algorithms.
- This validated approach offers a robust tool for quantitative analysis in clinical research and potentially in the diagnosis of AMD.
- The ability to distinguish between RPD and regular drusen automatically can enhance the understanding and management of nonneovascular AMD.
