Abstract
In the recent past, retinal image processing has become a popular biomedical modern computer aided diagnosis (CAD) system development technology for the detection and identification of eye related disease such as diabetic retinopathy, exudates, cardiovascular disease, glaucoma, etc. In present ophthalmology, the arrangement of retinal image with appropriate disease segmentation has attained greater attention for disease identification. Examining diameters of blood vessel inside vessels of retinal image during cardiac cycle (ECG gating) can assist cardiologists in the forecast of cardiovascular disease. Manual process of blood vessels on retinal image is a complex process due to risk handling of physician, noise occurrence on image and various types of acquisition execution. In this paper, we have proposed automatic diagnosis of cardiovascular disease using improved image processing methodologies using various CAD algorithms. The preprocessing steps are useful to improve the retinal image quality. The 2-Dimentional Adaptive Improved Bilateral Filter (2D-AIBF) is applied to remove noise interference on retinal image such as speckle noise, impulse noise and Gaussian noise. The contrast and brightness of retinal image is improved by applying Edge Preservation – Contrast Limited Adaptive Histogram Equalization (EP-CLAHE) algorithm. The blood vessel pixels are clustered by applying Arbitrary Robust Random Walks (ARRW) cluster algorithm. The Adaptive Otsu Threshold (AO) methodology is used to segment only Region Of Interest (ROI) blood vessels pixels and suppress other pixels. The Gray Level Co-Occurrence Matrix (GLCM) algorithm is used to extract the features on segmented image. The Deep Learning (DL) methodology is used to classify cardiovascular disease occurred or not. The Deep Convolutional Neural Network (CNN) is the type of DL technique that is applied for classification of cardiovascular disease. The experimental results are evaluated by comparing other conventional methods of retinal blood vessel segmentation with respect to accuracy, sensitivity and specificity using Confusion Matrix (CM) algorithm and the proposed methodology is proved to be more efficient and accurate in classification of cardiovascular disease classification.

Key words: 2D-AIBF, EP-CLAHE, ARRW, AO, GLCM, DCNN.