[Optic cup and disc segmentation model based on linear attention and dual attention]

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Context:

• Glaucoma is a leading cause of blindness, with the optic cup-to-disc ratio being a key diagnostic indicator.
• Accurate segmentation of the optic disc and cup is essential for reliable glaucoma screening.
• Existing segmentation methods may lack the precision required for early disease detection.

Purpose:

• To propose and evaluate a novel deep learning model for optic cup and disc segmentation.
• To leverage linear attention and dual attention mechanisms for enhanced feature extraction and calibration.
• To assess the model's performance on benchmark retinal image datasets.

Summary:

• A new segmentation model utilizes a linear attention ResNet-34 backbone for feature extraction.
• Channel and spatial dual attention mechanisms refine feature maps in the decoder for precise segmentation.
• The model achieved high Intersection over Union (IoU) scores on the DRISHTI-GS and RIM-ONE-V3 datasets.

Impact:

• The proposed model demonstrates superior performance compared to existing algorithms for optic cup and disc segmentation.
• This technology holds significant medical value for the early screening and diagnosis of glaucoma.
• Knowledge distillation is employed to create smaller, deployable models for embedded devices, enhancing accessibility.