An Intelligent system for early detection and diagnosis of diabetic retinopathy based on statistical and radon transform features of retina images

Abstract EN

The eye is one of the most important end organs for complications of diabetes.

There are two types of ocular diabetic complications : direct and indirect.

Diabetic retinopathy is the primary direct complication of diabetes in the eye.

There are two main categories of DR, Non- proliferative (Non-PDR) and Proliferative (PDR).

In this work a new technique is presented for automatic screening of diabetic Retinopathy based on Comprehensive Spectral-Textural features which are two fundamental patterns elements that can be interpreted by human.

To extract spectral features parameters (Bi-spectral invariant features at different angles of orientation) the gray level retina images are subjected to Radon transform which convert the 2D image in to ID image vector which introduced to the HOS (Higher Order Statistic) Scheme.

Nineteen Textural features parameter were derived from gray level matrix, co- occurrence matrix, and run length matrix features to perform texture analysis for retina images and hence to generate the features vector.

Later twenty-six best features form both categories (spectral & Textural) have been considered and introduced to the features selection stage to exclude irrelevant features.

Thus a feature vector is reduced to 10- relevant features for each retina image.

The database used is in this work is MESSADOR, with 150 retinal images have been used for training and validation stages .Two classifiers were presented, K-Nearest-Neighbor (KNN) and Naive Bayes Neural Network with different rates.

The simulation results show that the proposed system can attain very high Performance (Accuracy) about 98.09%.