•  
  •  
 

Abstract

Hypertensive Retinopathy (HR), a serious consequence of systemic hypertension, manifests through specific changes in the retinal vasculature observable in fundus images. There might not be any symptoms or complaints at the early stage of systemic hypertension. The complication of the systemic hypertension sets in vital organs of the body including the eyes over time. An automated method for HR detection can help in early detection and enhance diagnosis and management and thus reduce the risk of severe ocular and systemic complications. Subjective evaluation of the retinal images by ophthalmologist is the conventional diagnostic method, which might be inconsistent and time consuming. This paper presents an innovative deep neural network model AV-EffiCapsNet which integrates EfficientNet and Capsule Networks to detect HR in fundus images automatically. EfficientNet provides an efficient and scalable Conventional Neural Network framework and Capsule Networks enhance the representation of spatial hierarchies and part-whole relationships. The annotated fundus images of the datasets VICAVR and INSPIRE AVR were used to train and test the model AV-EffiCapsNet. The results showed superior precision of 97.7%, accuracy of 98.8% and recall of 95.5% compared to current models. These results indicate that AV-EffiCapsNet is effective in detecting subtle signs of HR, ensuring it a valuable tool for telemedicine and clinical screening.

Keywords

AV-EffiCapsNet, Artery-vein ratio analysis, Deep learning, Hypertensive retinopathy, Medical image analysis

Subject Area

Computer Science

Article Type

Article

First Page

3165

Last Page

3176

Creative Commons License

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

Download

Share

COinS