Automatic Screening And Analysis Study On Diabetic Retinopathy Images

Recent years with the improvement of living standards, the number ofdiabetes patients is growing rapidly. China has overtaken India as thecountry with the most diabetic patients in the world. Diabetic retinopathy isone of the commonest complications of diabetes, which is also the leadingcause of blindness of European countries in working age. Regularscreening is the only effective way to detect early lesions and conducttreatment before significant visual impairment. Fundus images are usuallyacquired in two ways: fundus fluorescence angiography and fundusphotography. We adopt fundus photography to obtain fundus images inorder to satisfy the screening. In this paper, we conduct relevant processingon diabetic retinopathy based on fundus photography.A new comprehensive enhancement method is proposed based on thecharacteristics of fundus images, which combines mathematicalmorphology, histogram equalization and two-dimensional matched filtersto improve image quality. Then, a new universal no-reference image qualityassessment method is introduced to evaluate our method objectively. Next,calculate the Hessian matrix of preprocessed images and multi-scale filters to extract vascular. Compare the results with experts’ manual segmentationresults to verify the effectiveness of our method. Finally, detect optic discprecisely according to its size, shape and gray scale.This paper conducted research on retinal image enhancement, vesselsegmentation and optic disc localication, which can be used for preliminaryscreening, clinical diagnosis and treatment result analysis. The system canaccomplish automatic detection and division stage on diabetic retinopathyby further functional improvement. Furthermore, a novel no-referenceimage quality assessment method is introduced to evaluate if the qualitybecome better or worse, which breakthrough the limits of traditionalevaluation methods.