Hybrid OCT Segmentation Algorithm for Retinal Layer Delineation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current OCT image segmentation methods are inadequate for accurately identifying and distinguishing between different retinal layers, such as the subretinal hyper reflective material, retina pigment epithelium, nerve fiber layer, and inner-outer retina interface, which are crucial for clinical diagnosis and treatment efficacy assessment.
Innovation Solution
A hybrid algorithm combining Bayesian Level Sets and graph-based methods is employed to segment OCT images, using preprocessing techniques like down-sampling and de-noising, and iterative graph weight calculations to identify and delineate these layers using Dijkstra's algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional segmentation methods are used for OCT images, then the processing is simpler, but the accuracy of identifying retinal layers is insufficient
Solution Approach 1:
The patent applies segmentation by dividing the OCT image processing into distinct stages: preprocessing (denoising, contrast enhancement), level set evolution for initial contour detection, and graph-based refinement for precise boundary identification. This multi-stage segmentation approach systematically improves measurement precision while managing algorithmic complexity through modular design.
Solution Approach 2:
The patent employs a composite algorithmic approach by combining level set methods with graph theory-based shortest path algorithms. This hybrid methodology integrates the continuous deformation capability of level sets with the discrete optimization strength of graph algorithms, achieving superior segmentation accuracy that neither method could achieve alone.
2Manufacturing precision
If advanced segmentation algorithms are used, then the segmentation precision is improved, but the computational complexity increases
Solution Approach 1:
The patent applies preliminary action through preprocessing steps (denoising, contrast enhancement) performed before the main segmentation algorithm. This prepares the image data to reduce computational burden during the intensive level set and graph-based processing stages, thereby improving boundary delineation precision while managing computational power requirements.
Solution Approach 2:
The patent implements partial action by applying graph-based shortest path refinement only to critical boundary regions identified by the level set method, rather than processing the entire image with the computationally intensive graph algorithm. This selective application maintains high boundary delineation precision while significantly reducing overall computational power consumption.
3Measurement precision
If iterative graph weight calculations are performed, then the layer identification accuracy is improved, but the processing time increases
Solution Approach 1:
The patent segments the iterative processing into distinct phases: initial level set evolution to establish candidate boundaries, followed by graph-based refinement iterations only in regions where layer identification is uncertain. This segmented approach maintains high layer identification accuracy while reducing total processing time by avoiding unnecessary iterations in already-well-defined regions.
Solution Approach 2:
The patent implements feedback mechanisms where segmentation results from previous iterations inform subsequent processing steps. The graph weight calculations use feedback from level set contours to focus computational effort on ambiguous regions, and refined boundaries from graph processing feed back to update level set initial conditions, creating an efficient adaptive loop that improves accuracy without linearly increasing processing time.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
The present disclosure describes a system and method to segment optical coherence tomography (OCT) images. The present system uses a hybrid method that employs both Bayesian level sets (BLS) and graph-based segmentation algorithms. The system first identifies retinal tissue within an OCT image using the BLS algorithms. The identified retinal tissue is then further segmented using the graph-based segmentation algorithms.