W-Net: A boundary-aware cascade network for robust and accurate optic disc segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate optic disc (OD) segmentation is crucial for diagnosing eye diseases.
• Variations in image acquisition lead to poor generalization in deep learning models.
• Existing methods struggle with diverse datasets, impacting diagnostic accuracy.

Purpose of the Study:

• To develop a robust multi-level segmentation network for accurate optic disc segmentation.
• To address the generalization challenges posed by varying image acquisition parameters.
• To enhance the performance of deep learning models in computer-aided ophthalmic diagnosis.

Main Methods:

• A novel multi-level segmentation network incorporating a data quality enhancement module (DQEM), coarse segmentation module (CSM), and localization module (OLM).
• Introduction of W-Net within the fine segmentation stage module (FSM) for the first time.
• Integration of boundary loss into the loss function to refine segmentation accuracy.

Main Results:

• The proposed network achieved superior optic disc segmentation performance across multiple benchmark datasets (REFUGE, GAMMA, Drishti-GS1, IDRiD).
• The multi-level approach effectively handled variations in image resolution, size, contrast, and clarity.
• W-Net and boundary loss significantly improved segmentation accuracy and generalization capabilities.

Conclusions:

• The developed multi-level segmentation network offers a significant advancement in optic disc segmentation accuracy.
• The method demonstrates robust generalization, outperforming state-of-the-art networks on diverse ophthalmic image datasets.
• This approach holds promise for improving computer-aided diagnosis systems for various eye conditions.