Automatic Segmentation in Multiple OCT Layers For Stargardt Disease Characterization Via Deep Learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Stargardt disease is a leading inherited macular dystrophy.
• Accurate segmentation of retinal layers in spectral-domain optical coherence tomography (SD-OCT) is crucial for understanding disease progression.
• Current manual segmentation methods are time-consuming and subjective.

Purpose of the Study:

• To develop and validate an automated algorithm for segmenting 11 retinal layers and Stargardt-associated features in SD-OCT images.
• To analyze morphologic differences between normal eyes and those with Stargardt disease.
• To establish a quantitative tool for Stargardt disease monitoring.

Main Methods:

• Implementation of a deep learning-shortest path (DL-SP) framework, combining deep learning with shortest path segmentation.
• Generation of retinal layer thickness and intensity feature maps.
• Comparison of automated segmentation with manual tracings by expert graders.

Main Results:

• The DL-SP algorithm achieved subpixel accuracy for all retinal layers compared to manual segmentation.
• Accurate identification of Stargardt disease features, including flecks and atrophic lesions.
• Demonstrated visualization of characteristic Stargardt morphologic changes.

Conclusions:

• This is the first automated algorithm for segmenting 11 retinal layers in SD-OCT for Stargardt disease.
• The identified feature differences offer new insights into Stargardt disease pathophysiology.
• The algorithm supports quantitative monitoring of Stargardt disease using SD-OCT.