Retinal layer and fluid segmentation in optical coherence tomography images using a hierarchical framework.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of retinal layers and fluid in optical coherence tomography (OCT) images is crucial for diagnosing and monitoring retinal diseases.
• Existing joint segmentation methods show promise for retinal layers but struggle with satisfactory fluid segmentation.

Purpose of the Study:

• To develop an improved hierarchical framework for precise segmentation of retinal layers and fluid in OCT images.
• To address the limitations of current methods in accurately segmenting fluid-filled regions.

Main Methods:

• A sequential training approach using three fully convolutional networks.
• A novel weighting scheme for the loss function, informed by previously trained networks.
• Modification of the mutex Dice loss to penalize errors between distant retinal layers, enhancing positional accuracy.

Main Results:

• The hierarchical framework achieved superior segmentation performance for the inner segment ellipsoid layer (Dice coefficient = 0.95) and fluid (Dice coefficient = 0.82).
• Segmentation results for other retinal layers reached state-of-the-art levels.
• Significant improvements in fluid segmentation were observed without compromising retinal layer segmentation.

Conclusions:

• The proposed hierarchical framework offers substantial advancements in segmenting retinal fluid and layers from OCT images.
• This technology can serve as a valuable tool for ophthalmologists, acting as a second opinion or enabling automated biomarker extraction.