Diabetic retinopathy (DR) is the major cause of visual impairment among diabetic patients. Significant works have been done to hybrid a modified CNN architecture such as AlexNet with some of classifiers such as support vector machines (SVMs) or fuzzy C-Means (FCM) to improve the DR screening. This new hybrid innovative structure uses more efficient extracting features of a retinal images in both spatial and spectral domains. In spite the advantages of this innovative architecture, the different kernel functions affect the performance of the proposed algorithm. By using the appropriate transformed data into two or three dimensional feature maps and using an improved support vector domain description (ISVDD) can obtain more flexible and more accurate image description. To this end, the optimal degree values of different kernel functions can be extracted by using a particle swarm optimization (PSO) algorithm. Also, we compared the performance of our approach (modified-AlexNet-ISVDD) with the results obtained by hybrid modified AlexNet and some of classifiers such as K-Nearest Neighbors (KNN) and FCM clustering. We achieve the proposed CNN architecture using ISVDD on the DIARETDB1 and MESSIDOR datasets, with more than 99% sensitivity.

Keywords