Advanced OCTA imaging segmentation: Unsupervised, non-linear retinal vessel detection using modified self-organizing...

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of the retinal vascular system is crucial for diagnosing eye diseases.
• Optical Coherence Tomography Angiography (OCTA) provides high-resolution images of retinal vasculature.
• Developing automated methods for OCTA image analysis is essential for efficient Computer-Aided Diagnosis (CAD).

Purpose of the Study:

• To propose a fully automated and unsupervised stochastic segmentation approach for retinal vascular system detection in OCTA images.
• To leverage a two-level joint Markov-Gibbs Random Field (MGRF) model for enhanced segmentation accuracy.
• To facilitate the development of CAD systems for retinal disease detection.

Main Methods:

• A novel probabilistic model based on Linear Combination of Discrete Gaussian (LCDG) was used to model OCTA image appearance.
• A modified Expectation Maximization (EM) algorithm estimated LCDG model parameters.
• A two-level joint MGRF model, with analytically estimated parameters and modified self-organizing maps, performed unsupervised segmentation.

Main Results:

• The proposed segmentation framework achieved a Dice similarity coefficient of 0.92 ± 0.03.
• A 95-percentile bidirectional Hausdorff distance of 0.69 ± 0.25 was obtained.
• An accuracy of 0.93 ± 0.03 confirmed the superior performance of the approach.

Conclusions:

• The developed unsupervised and fully automated segmentation approach demonstrates superior performance in detecting the vascular system from OCTA images.
• This method offers a significant advancement over traditional techniques, providing more accurate segmentation results.
• The approach holds substantial potential for improving CAD systems in the detection of retinal diseases.