Automatic grading of retinal blood vessel tortuosity using Modified CNN in deep retinal image diagnosis. (April 2022)

• Record Type: Journal Article
• Authors: Maji, Debasis; Maiti, Souvik; Dhara, Ashis Kumar; Sarkar, Gautam
• Is Part Of: Biomedical signal processing and control. Volume 74(2022)
• Publication Date: 2022-04
• Subjects: Diabetic retinopathy (DR) -- Convolutional Neural Network (CNN) -- Fundus image -- Retinal Blood Vessel; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103514
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Background: The World Health Organization states that the number of patients suffering from diabetes has shot up by nearly four times from 108 millions in 1980 to 422 millions in 2014. Diabetic Retinopathy (DR) is the long-term effect of diabetes, which if not clinically treated effectively on time might lead to irreversible loss of vision. By examining the retinal fundus images the disease might be diagnosed well by examining the retinal fundus images. However, the fact that these images contain noise and variation due to certain environmental conditions such as light makes it difficult even for experts in the field to access the right grade of the disease. In this paper, we aim to present a robust Convolution Neural Network (CNN) architecture, which can grade the disease irrespective of noise and variation. We have used the publicly available dataset on Kaggle to train our model and we have validated on another publicly available data set, EIARG2. We also provide a comparative study of our model against standard architectures like ResNet50, VGG16 and several others of the domain and thus conclude by virtue of promising results that our architecture is superior for grading diabetic retinopathy than the present-day standard architectures. A CNN network has been proposed which can grade retinal images using the state-of-the-art machine learning algorithms. The method found 0.96 Performance (SCC) accurate for grading the tortuosity-based eye health.