Multi-phase level set algorithm based on fully convolutional networks (FCN-MLS) for retinal layer segmentation in...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Retinal layer thickness is crucial for diagnosing retinal diseases.
• Manual segmentation is time-consuming and prone to bias.
• Automatic segmentation is challenging due to image noise and complex retinal structures.

Purpose of the Study:

• To develop an accurate automatic segmentation method for retinal layers in spectral OCT images.
• To address challenges like speckle noise, low contrast, and irregular morphology.
• To segment nine retinal boundaries for improved clinical information extraction.

Main Methods:

• Proposed a coarse-fine framework combining a full convolutional network (FCN) with a multiphase level set (MLS).
• FCN performs initial classification of retinal layers and boundary extraction.
• MLS refines segmentation using regional restrictions for accurate nine-boundary identification.

Main Results:

• The FCN-MLS method accurately segmented nine retinal boundaries in 1280 B-scans.
• Achieved high consistency with manual segmentation (mean absolute boundary difference: 5.88 ± 2.38μm).
• Outperformed existing state-of-the-art methods in segmentation accuracy.

Conclusions:

• The proposed FCN-MLS framework offers a reliable and accurate solution for automatic retinal layer segmentation.
• This method can significantly aid in the diagnosis and monitoring of retinal diseases.
• The approach demonstrates superior performance compared to current state-of-the-art techniques.