Choroidal Layer Analysis in OCT images via Ambiguous Boundary-aware Attention

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) visualizes choroidal morphology but struggles with sublayer and vessel segmentation due to indistinct boundaries and imbalanced distributions.
• Quantitative analysis of choroidal sublayers and vessels in OCT images remains underexplored.

Purpose of the Study:

• To propose a novel two-stage deep learning architecture, the Choroidal Layer Analysis network (CLA), for joint segmentation of choroidal sublayers and vessels in OCT images.
• To address challenges of indistinct boundaries and imbalanced vessel distribution for improved segmentation accuracy.

Main Methods:

• Developed a two-stage architecture (CLA) using an encoder-decoder backbone with residual U-shape modules.
• Incorporated an Ambiguous Boundary Attention (ABA) block in the first stage to enhance segmentation of subtle choroidal sublayer boundaries.
• Utilized an active contour-based loss in the second stage for refining choroidal vessel contours and contextual modeling.

Main Results:

• The CLA network achieved superior performance compared to state-of-the-art segmentation methods on a dataset of 800 annotated OCT images.
• Successfully reconstructed large choroidal vessels in 3D and calculated morphological parameters.
• Demonstrated significant differences in 3D parameters between healthy controls and high myopia groups.

Conclusions:

• The proposed CLA network offers a robust solution for joint choroidal sublayer and vessel segmentation in OCT images.
• The 3D morphological analysis provides valuable insights for understanding diseases like high myopia.
• This work facilitates clinical decision-making and disease understanding through advanced OCT image analysis.