A Deep-Learning Approach for Automated OCT En-Face Retinal Vessel Segmentation in Cases of Optic Disc Swelling Using...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optic disc swelling presents challenges for retinal blood vessel segmentation in optical coherence tomography (OCT) volumes due to shadowing artifacts.
• Accurate vessel segmentation is crucial for assessing the severity and cause of optic disc swelling.

Purpose of the Study:

• To develop a deep-learning-based approach for enhanced retinal blood vessel segmentation in OCT volumes, specifically addressing challenges posed by optic disc swelling.
• To investigate the hypothesis that combining vessel information from multiple projected retinal layers improves vessel visibility.

Main Methods:

• A deep neural network was trained using three OCT en-face images (retinal pigment epithelium, inner retina, total retina) as input.
• The model was trained on data from 18 patients with optic disc swelling.
• A human expert vessel tracing, incorporating fundus images, served as the reference standard.

Main Results:

• The proposed deep learning approach demonstrated superior performance compared to manual tracing methods on an independent test set of 18 patients.
• Manual tracing utilizing three OCT en-face images outperformed traditional single RPE en-face image tracing.
• The study highlights the benefit of multi-layer OCT image integration for vessel segmentation.

Conclusions:

• Simultaneous consideration of vessel information from multiple projected retinal layers significantly enhances vessel visibility in OCT scans of optic disc swelling.
• The proposed method offers improved vessel segmentation compared to traditional single-image approaches.
• This advancement can lead to more accurate assessments of optic disc swelling severity and etiology.