Structure-Constrained Regression Network for Efficient and Topology-Guaranteed Retinal Layer Segmentation in OCT...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate retinal layer segmentation in OCT images is crucial for diagnosing and monitoring eye diseases.
• Challenges include speckle noise, intensity variations, and pathologies that impede segmentation.
• Topology-guided segmentation relies on structured layer boundaries, which are often disrupted by interference factors.

Purpose of the Study:

• To present a novel, efficient, and topology-guided deep learning network for retinal layer segmentation in OCT images.
• To leverage inherent structural priors of retinal layers for robust boundary segmentation.
• To improve the accuracy and efficiency of retinal layer segmentation despite interference factors.

Main Methods:

• Developed a lightweight, end-to-end deep network named Structure-Constrained Regression Network (SCRNet).
• Employed a two-stream architecture to capture layer topology and boundary continuity separately.
• Integrated tailored Structural Feature Modules (SFMs) and Structure-Constrained Losses (SCLs) within each stream.
• Utilized a Structure-Constrained Regression Module (SCRM) to integrate information from both streams for enhanced boundary regression.

Main Results:

• SCRNet achieved state-of-the-art performance in segmenting structured layer boundaries and topology-guaranteed layers.
• Demonstrated high efficiency in segmentation tasks.
• Validated performance on two publicly available benchmark datasets.

Conclusions:

• SCRNet effectively leverages structural priors for accurate and robust retinal layer segmentation in OCT images.
• The proposed method overcomes common interference factors, enabling reliable quantification of retinal morphology.
• SCRNet offers a promising tool for ophthalmic disease diagnosis and monitoring.