EARDS: EfficientNet and attention-based residual depth-wise separable convolution for joint OD and OC segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma is a leading cause of irreversible vision loss.
• Accurate segmentation of the optic disc (OD) and optic cup (OC) is crucial for glaucoma diagnosis.
• Optic cup segmentation presents challenges due to shape variability and indistinct boundaries, often degrading deep learning model performance.

Purpose of the Study:

• To propose an efficient one-stage network for joint OD and OC segmentation.
• To address the limitations of independent segmentation and pre-processing requirements.
• To enhance the accuracy and efficiency of glaucoma diagnostic imaging analysis.

Main Methods:

• Developed EfficientNet and Attention-based Residual Depth-wise Separable Convolution (EARDS) network.
• Utilized EfficientNet-b0 as an encoder for robust boundary representation.
• Incorporated Attention Gate (AG) and Residual Depth-wise Separable Convolution (RDSC) blocks in a novel decoder for improved feature highlighting and efficiency.
• Employed a weighted combination of focal loss and dice loss for precise segmentation guidance.

Main Results:

• The proposed EARDS model demonstrated superior performance compared to state-of-the-art methods on Drishti-GS and REFUGE datasets.
• The network achieved accurate joint segmentation of OD and OC.
• The architecture improved computational efficiency and addressed challenges in OC segmentation.

Conclusions:

• EARDS offers an effective solution for joint OD and OC segmentation in glaucoma diagnosis.
• The model's design enhances segmentation accuracy and computational efficiency.
• This approach contributes to advancing automated glaucoma detection through improved medical image analysis.