Automatic Screening of Diabetic Retinopathy Using Fundus Images and Machine Learning Algorithms

Abstract

Diabetic Retinopathy is a vision impairment caused by blood vessel degeneration in the retina. It is becoming more widespread as it is linked to diabetes. Diabetic retinopathy can lead to blindness. Early detection of diabetic retinopathy by an ophthalmologist can help avoid vision loss and other complications. Diabetic retinopathy is currently diagnosed by visually recognizing irregularities on fundus pictures. This procedure, however, necessitates the use of ophthalmic imaging technologies to acquire fundus images as well as a detailed visual analysis of the stored photos, resulting in a costly and time-consuming diagnosis. The fundamental goal of this project is to create an easy-to-use machine learning model tool that can accurately predict diabetic retinopathy using pre-recorded digital fundus images. To create the suggested classifier model, we gathered annotated fundus images from publicly accessible data repositories and used two machine learning methods, support vector machine (SVM) and deep neural network (DNN). On test data, the proposed SVM model had a mean area under the receiver operating characteristic curve (AUC) of 97.11%, whereas the DNN model had a mean area AUC of 99.15%.

Classification of Diabetic Retinopathy Images through Deep Learning Models - Color Channel Approach: A Review

“…ISSN 2549-7286 (online) Thus, from the above observation of an instance of experimentation, the value of parameter 'p' is still corrected to p = σx ± SEM, where, if the confidence is considered as 95% of samples possess exudates p1 = σx + SEM and for less than 95% samples σx = sd -SEM. Probability of success always lies between 90% to 99%, optimistically the heuristic is to select the 95%, this is by mimicking the feature heuristic used in [76][77] [78] and by Yuen et al The 95% is not a fixed value for the confidence, a heuristically selected value which represents the highest of the percentage probabilities in the scale of 1 to 100.…”

A Low-Cost Diabetic Retinopathy Screening Tool Using a Smartphone and Machine Learning Algorithm

“…. GLCM, denoted by may be de ned as a square matrix of size N X N, where N represents the total number of grey levels in the image, with each of its elements give number of times a speci c grey levels co-exists in the image along certain orientations, d = {0, 45, 90, 135, …}, at a particular distances, k = {1, 2, 3,4…}, where m and n denotes the intensity levels of the pixel of interest [19]. The computation of GLCM for a given image patch of size 4x4 having four intensity levels (0, 1, 2, 3) is illustrated in Fig.…”

“…We then computed ten GLCM features that include mean, SD, energy, contrast, entropy, homogeneity, correlation, RMS, skewness, and kurtosis, which are given in Eqs. 4-14 [19,20].…”

“…Most recently, ten GLCM features were gathered by another team of researchers who used them to create two binary classi ers, a support vector machine (SVM) and a deep neural network (DNN), to categorize pre-recorded retinal pictures as healthy or DR. The SVM and DNN both had cross-validated classi cation accuracy scores of 94.55% and 95.77%, respectively [9].…”

Detection Diabetic Retinopathy With Supervised Learning

“…NPDR is the early stage of disease progression, while PDR is a more advanced and prolonged phase. In PDR, patients may experience the growth of new blood vessels in the eye, which may eventually lead to blindness [1].…”