ECSD-Net: A joint optic disc and cup segmentation and glaucoma classification network based on unsupervised domain...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma can lead to irreversible vision loss and blindness.
• Early diagnosis is crucial for preventing vision loss.
• Analyzing optic disc and cup is key for glaucoma diagnosis.

Purpose of the Study:

• Develop an unsupervised learning method to address dataset distribution gaps in glaucoma diagnosis.
• Improve the prediction performance of deep learning models on new datasets.
• Enhance the generalization and efficiency of computer-aided glaucoma diagnosis.

Main Methods:

• Proposed a novel unsupervised model using adversarial learning for optic disc and cup segmentation and glaucoma screening.
• Employed unsupervised domain adaptation on the output space of a segmentation network to mitigate domain shift.
• Combined classification and segmentation networks for stable and efficient glaucoma screening predictions.

Main Results:

• Effectively alleviated performance degradation caused by domain shift in segmentation.
• Improved the accuracy of glaucoma screening.
• Outperformed state-of-the-art unsupervised domain adaptation methods for optic disc and cup segmentation.

Conclusions:

• The proposed method assists clinicians in glaucoma screening and diagnosis.
• The model is suitable for real-world clinical applications.
• Offers a generalized and efficient approach to computer-aided glaucoma diagnosis.