The Research Of The Diabetic Retinopathy Detection System Based On Fundus Images

Diabetic retinopathy is one of the most serious diabetes complications, but also present the main blinding factor in the working age population. Regular fundus examination is the key to early detection and timely treatment before patients’ visual injury. It is very necessary to develop a retinopathy detection system which would help millions of people living in areas of poor medical conditions.On account of the poor quality fundus images would make it difficult in fundus image analysis and medical diagnostic, we evaluate whether the image suitable for diagnostic purposes by applying quality evaluation algorithm to the image obtained. The algorithm effectiveness is thoroughly appraised through a large amount of public datasets, and obtained subjective evaluation results consistent with human visual perception by statistical analysis of retinal images. Then, based on the characteristics of the fundus image, an enhanced algorithm of fundus image enhancement is proposed, combined with HSI color space conversion, median filtering and contrast limited adaptive histogram equalization to improve the quality of fundus images. Then, converting the pre-processed image to Lab space which has wider color gamut, and using fast K-means clustering algorithm for coarse segmentation. Finally, we extract hard exudates from the coarse segmentation candidate region with support vector machine.Based on the research of the diabetic retinopathy detection system, gives some reference fundus image acquisition scheme. In addition, a common fundus image quality evaluation method which overcome the problem of high computation cost and single indicator compared with traditional method is proposed in this paper. The new algorithm has high efficiency and strong practicability. Furthermore, the calculated indicators of general quality are very useful for retinal image acquisition device parameter adjustment. Moreover, this paper implements retinal image enhancement, optic disc localization and hard exudates detection, which can realize the preliminary analysis and the diagnosis to the diabetes retinopathy, and has the very big help to the doctor’s diagnosis and the surgery analysis. At last, we verify the feasibility of this scheme by building a experimental platform for retinopathy detection system using ophthalmoscope of 66 vision Inc and a smartphone.