disease detection from retinal images using deep learning techniques

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: disease detection from retinal images using deep learning techniques

ارائه دهنده: Provider: Siamand Avestan

اساتید راهنما: Supervisors: Hassan Khotanlou (Ph.D)

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: Mir Hossein Dezfoulian (Ph.D) and Muharram Mansoorizadeh (Ph.D)

زمان و تاریخ ارائه: Time and date of presentation: 2021/10/20

مکان ارائه: Place of presentation: Engineering College

چکیده: Abstract: "Symptoms of many diseases, including diabetes retinopathy, appear in the early stages of the retina. Diabetic retinopathy is a common complication of diabetes and one of the most important causes of blindness. Many of the complications of diabetes retinopathy can be prevented with early diagnosis. Therefore, early detection of the disease is very important for successful treatment. Today, due to the increasing volume of medical images, there is a need for many specialists to interpret them, a need that is not available everywhere. Simplifying the diagnosis is very important and can help millions of people avoid the complications or blindness of retinopathy of diabetes. Therefore, retinal examination by automated methods plays an important role in the early diagnosis of retinopathy of diabetes. Recent modeling in the field of artificial intelligence in medical image processing includes models based on deep convolutional neural networks. In this study, a method based on deep convolutional neural network for the problem of diagnosis and classification of diabetic retinopathy in retinal Fondus images is presented. This research generally consists of two parts: preprocessing and classification. The preprocessing section includes methods for removing redundant pixels from the image and using enhanced techniques. The categorization section consists of three categories based on deep convolutional neural networks, each of which has a good performance and categorizes images with imaginative accuracy. The first category is a deep neural network consisting of continuous convolutional blocks. The second network consists of two parallel networks with the same filters and different kernels. The third network is the first network that is used hierarchically. Classification of retinal images, in this study, according to the classes of images, is used in several ways. But the most important and general form of classification in this study is the classification into five class of healthy people and people with mild, moderate, severe and proliferative diabetes retinopathy. Finally, the proposed method is applied and tested on Aptos dataset. The proposed method is tested on this dataset using appropriate criteria. The result of comparing the proposed method with the previous methods is acceptable and has left a good performance model."

فایل: ّFile: Download فایل