Structured layer surface segmentation for retina OCT using fully convolutional regression networks

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Optical coherence tomography (OCT) is crucial for retinal imaging, with retinal layers serving as key disease biomarkers.
• Accurate segmentation of retinal layers is essential for automated retinal thickness and surface shape analysis.
• Current state-of-the-art methods employ a two-step process: pixel classification followed by graph-based surface extraction, which has limitations.

Purpose of the Study:

• To develop a unified deep learning framework for direct, single-step segmentation of smooth, continuous, and topologically correct retinal layer surfaces.
• To overcome the limitations of current two-step methods in extracting structured surfaces with topological constraints.

Main Methods:

• A novel deep learning framework was proposed, directly modeling the distribution of surface positions.
• This unified approach combines pixel classification and surface extraction into a single feed-forward operation.
• The method was evaluated on public datasets including healthy controls and patients with multiple sclerosis or diabetic macular edema.

Main Results:

• The proposed method successfully achieved smooth, continuous, and topologically correct retinal layer surface segmentation.
• It demonstrated state-of-the-art performance with sub-pixel accuracy on diverse datasets.
• The unified framework significantly improved the efficiency and accuracy of retinal layer segmentation.

Conclusions:

• The developed deep learning framework offers a significant advancement in automated retinal layer segmentation using OCT.
• This method provides a more efficient and accurate approach for analyzing retinal thickness and surface shape, aiding in disease diagnosis.
• The unified, data-driven approach sets a new benchmark for OCT image analysis in ophthalmology.