PGKD-Net: Prior-guided and Knowledge Diffusive Network for Choroid Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Choroidal thickness is a key diagnostic indicator in ophthalmic conditions.
• Accurate segmentation of the choroid layer in Optical Coherence Tomography (OCT) images is essential for disease monitoring.
• Challenges in choroid segmentation include blurry boundaries and interference from lesions.

Purpose of the Study:

• To develop a novel deep learning network for accurate choroid segmentation in retinal OCT images.
• To enhance the utilization of retinal structural information for improved segmentation performance.
• To address the limitations of existing methods in handling blurry boundaries and lesion interference.

Main Methods:

• Proposed a Prior-mask Guided and Knowledge Diffusive Network (PGKD-Net).
• The network comprises a Prior-mask Guided Network (PG-Net) for initial segmentation and a Knowledge Diffusive Network (KD-Net) for refinement.
• Introduced Multi-Scale Context Aggregation (MSCA) and Multi-Level Feature Fusion (MLFF) modules for feature enhancement.

Main Results:

• The PGKD-Net effectively utilizes retinal structural information to highlight choroidal features.
• MSCA module captures long-distance dependencies for improved global context learning.
• MLFF module integrates contextual knowledge for enhanced fine-level segmentation.
• Experimental results demonstrate superior segmentation accuracy compared to state-of-the-art methods.

Conclusions:

• The proposed PGKD-Net achieves high accuracy in choroid segmentation from retinal OCT images.
• The novel network architecture and feature enhancement modules effectively overcome segmentation challenges.
• This method holds significant potential for clinical diagnosis and monitoring of ophthalmic diseases.