Robust total retina thickness segmentation in optical coherence tomography images using convolutional neural networks

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate retina segmentation is crucial for diagnosing and monitoring eye diseases using optical coherence tomography (OCT).
• Existing segmentation methods struggle with severe retinal pathologies, leading to inaccurate measurements.

Purpose of the Study:

• To develop a fully automated and robust system for total retina segmentation in OCT images.
• To improve the accuracy of macular thickness estimation, especially in cases with significant retinal abnormalities.

Main Methods:

• Development of a novel system employing a generalized U-net convolutional neural network (CNN) architecture.
• The CNN was designed to capture large contextual information to handle extensive retinal variations.
• The system was evaluated against two existing algorithms for segmentation performance and macular thickness accuracy.

Main Results:

• The proposed algorithm demonstrated superior qualitative and quantitative performance compared to existing methods.
• Achieved a macular thickness estimation error of 14.0 ± 22.1 µm, significantly lower than competitors (42.9 ± 116.0 µm and 27.1 ± 69.3 µm).
• The system proved robust in segmenting retinas with severe pathological changes.

Conclusions:

• The developed automated CNN system provides reliable retina segmentation, even in the presence of severe pathology.
• This advancement offers a more accurate tool for assessing macular thickness in diverse OCT imaging scenarios.
• The generalized U-net architecture effectively models the wide variability of retinal appearances.