Joint optic disc and cup segmentation based on densely connected depthwise separable convolution deep network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Glaucoma is a leading cause of vision loss, necessitating early detection through indicators like the cup to disc ratio (CDR).
• Accurate segmentation of the optic disc and optic cup is crucial for calculating the CDR, a key diagnostic metric.
• Existing deep learning methods struggle with precise segmentation due to significant overlap between the optic disc and cup in fundus images.

Purpose of the Study:

• To develop an advanced deep learning model for accurate joint segmentation of the optic disc and optic cup in fundus images.
• To improve the accuracy of the cup to disc ratio (CDR) calculation for enhanced glaucoma screening and diagnosis.

Main Methods:

• A two-stage approach is proposed: initial optic disc localization followed by joint segmentation of the optic disc and cup.
• A novel deep learning model, DDSC-Net (densely connected depthwise separable convolution network), is introduced for multi-category semantic segmentation.
• The model utilizes depthwise separable convolutional layers and an image pyramid input to create a deeper and wider network architecture.

Main Results:

• DDSC-Net achieved high segmentation accuracy on the Drishti-GS and REFUGE datasets, outperforming state-of-the-art methods.
• Disc coefficient scores of 0.9780 (optic disc) and 0.9123 (optic cup) were obtained on the Drishti-GS dataset.
• The method demonstrated superior performance in optic cup segmentation, outperforming GL-Net and pOSAL on both datasets.

Conclusions:

• The proposed DDSC-Net model shows significant potential for assisting in the screening and diagnosis of glaucoma.
• The enhanced segmentation accuracy of the optic disc and cup can lead to more reliable CDR measurements.
• This method offers a promising advancement in automated glaucoma detection systems.