پيشبيني بيماران ديابتي از آنژيوگرافي مبتني بر روش رشد ناحيه با استفاده از الگوريتم تركيبي GA - FCM

Diabetic retinopathy or morphological lesions are associated with abnormalities in retinal blood flow. These lesions indicate a regional distribution that includes risk factors in the early stages of the disease and can predict disease progression. The development of diabetic retinopathy is also associated with regional changes in retinal blood flow and the regulation of retinal capillary diameter in the macular area and in the retinal environment. Although diabetic retinopathy is common in the world today, it is difficult for people to prevent. The ophthalmologist usually diagnoses the severity of retinopathy of the eye through a visual examination of the fundus by direct examination and eva‎luation of color photographs. Due to the large number of diabetic patients around the world, this process is expensive and costly. The aim of this study is to provide an unsupervised computational diagnostic method for the diabetic diagnosis of angiographic images of human eye. To achieve this goal, we used the growth region method to identify the area. In the growth region method, based on the similarity or homogeneity of adjacent pixels, the image is divided into separate regions according to the criteria used for homogeneous analysis to determine their similarity to the relevant area. In this research, two manual methods and the use of FCM as fitness function of the genetic algorithm, are used. The algorithm was performed for 100 healthy eyes and 100 patients with retinopathy. The results show that the GA-FCM method works better than the manual method for selecting initial seeds. The results of comparing the fuzzy fit function in the genetic algorithm with other techniques show that the proposed model performed better with the highest value of the jacquard index and the shortest jacquard distance than the manual method. The accuracy of the proposed method is 87.5%, which is better than the manual method with 78.5% accuracy. Also, the sensitivity of the proposed method in the detection of the target area is 92%, which offers better performance than the manual method with 85% sensitivity.

1401/02/18

سميه طاهري

69678