•  
  •  
 

Corresponding Author

akanksha bali

Subject Area

Biomedical Engineering

Article Type

Original Study

Abstract

The anatomy of the retinal fundus picture, which includes the optic disc, affects how glaucoma characteristics are extracted. One of the main causes of blindness in the globe is glaucoma. Correct glaucoma screening in its early phases is difficult since the condition only becomes apparent when its symptoms are severe. If glaucoma is not treated, the optic nerve head (ONH) will be harmed, resulting in vision loss. Nevertheless, there are currently not enough eye experts accessible, and eye screening is personal, time-consuming and labor-intensive. Therefore, this is a real issue that can be addressed by automatically diagnosing glaucoma using deep learning (DL) algorithms. Routine glaucoma screening is thus recommended and required. This necessitates the use of an automated segmentation system that can precisely identify the areas with margins and help ophthalmologists monitor and diagnose the severity of glaucoma early on. A hybrid density-ED-UHI encoder-decoder-based unet hybrid inception model with 15-fold cross-validation is presented in the recommended research for the diagnosis of glaucoma. The proposed model was trained and validated using the REFUGE (Retinal Fundus Glaucoma Challenge) dataset. For the segmentation of glaucoma, the model achieves a 99.9% accuracy, 97.2% auc, 99.9% sensitivity, and 93.7% specificity. This remarkable outcome implies that fundus pictures' blood vessel segmentation may be used as a substitute for automatically detecting glaucoma.

Keywords

Glaucoma, REFUGE, InceptionV3, UNet, ResNet18, ResNet34, Segmentation.

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS