A generative model for OCT retinal layer segmentation by integrating graph-based multi-surface searching and image...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of retinal layers in Optical Coherence Tomography (OCT) is crucial for diagnosing and monitoring eye diseases.
• Existing automated segmentation methods often struggle with complex layer structures and inter-subject variability.

Purpose of the Study:

• To propose a novel generative probabilistic modeling framework for automated segmentation of retinal layers from OCT data.
• To develop a machine learning approach that learns a segmentation protocol from manually labeled training images.

Main Methods:

• Integration of simultaneous searching of multiple interacting layer interfaces, image registration, and machine learning.
• Incorporation of spatial constraints on retinal layer layout, robust local image descriptors, and inter-subject anatomical similarities.
• Training the model using manually labeled OCT volumetric images.

Main Results:

• The proposed model achieved superior performance in segmenting retinal layers compared to two state-of-the-art techniques.
• Experimental validation was conducted on OCT volumetric images from mutant canine retinas.

Conclusions:

• The developed generative probabilistic modeling framework offers a novel and effective approach for OCT retinal layer segmentation.
• The method's ability to integrate spatial constraints, robust descriptors, and anatomical similarities enhances segmentation accuracy.