Identification of Diabetic Retinopathy from Color Fundus Images using Deep Convolutional Neural Network
Bansode Balhim Narhari1, Bakwad K.M.2, Ajij Dildar Sayyad3

1Bansode Balbhim Narhari, Research scholar M.I.T Aurangabad, India.
2Bakwad K.M., Principal Puranmal Lahoti Government, Polytechnic college, Latur, MSBTE, Mumbai, India.
3Ajij Dildar Sayyad, Associate Professor and Vice Principal,MIT Aurangabad, India.

Manuscript received on April 02, 2020. | Revised Manuscript received on April 21, 2020. | Manuscript published on May 30, 2020. | PP: 878-884| Volume-9 Issue-1, May 2020. | Retrieval Number: F9905038620/2020©BEIESP | DOI: 10.35940/ijrte.F9905.059120
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: Existing methods on retinal disease detection are mostly depends on lesion detection techniques or multiple instance learning framework. However extensive research efforts fail to address effective representations of the different lesions from fundus images. In this paper, a innovative techniques is offered built on pre-examined entirely convolutional neural network (CNN) over and done with transfer learning. The proposed method utilizes the effective learning from recent deep CNN models with use of SVM classifier at the end. Meanwhile, additional retinal image pre-processing technique is applied for the better classification results. The improved result has contributed to the area of computer aided diagnosis for retinal screening system. Extensive experiments have been conducted on Messidor and IDRiD database with desired obtained accuracy of 96.29 % and 94.82 %. The proposed method supports retinal disease screening effectively by deep learning methods. 
Keywords:  Diabetic retinopathy (DR), Deep convolutional neural network, deep learning, transfer learning, pre-trained model.
Scope of the Article: Deep learning