3DChoroidSwin: advancing 3D choroid segmentation in OCT images through Swin Transformer and morphological guidance

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• The choroid, a vascular layer crucial for retinal health, is increasingly studied using optical coherence tomography (OCT).
• Accurate choroidal segmentation in OCT images is vital for understanding eye diseases, but current methods struggle with 3D structures and long-range dependencies.

Purpose of the Study:

• To develop an advanced automatic method for segmenting the choroid in 3D OCT images.
• To overcome limitations of existing 2D and CNN-based segmentation techniques.

Main Methods:

• Proposed 3DChoroidSwin method combining 3D CNN and 3D Swin Transformer for comprehensive feature learning.
• Incorporated a boundary loss function for shape-aware training and reduced false positives.

Main Results:

• 3DChoroidSwin achieved superior performance compared to existing methods on clinical OCT data.
• The method demonstrated high accuracy, comparable to ground truth segmentation.
• Generated smooth, continuous 3D choroidal segmentations with minimized errors at the margins.

Conclusions:

• 3DChoroidSwin offers a robust and accurate solution for 3D choroidal segmentation in OCT images.
• The method holds significant potential for improving the diagnosis and monitoring of various eye conditions.