Skip to content

JUSTINE-A1107/Diabetic-Retinopathy-by-eye-segmentation-with-eye-detection

Repository files navigation

Diabetic-Retinopathy-by-eye-segmentation-with-Eye-detection

MATLAB-Image Processing We have proposed a processing tool which will be helpful in identifying the disorder with maximum accuracy. The system applies filtering techniques and detects the diabetic retinopathy defect. Based on the severity of the defect, it is identified whether it’s in proliferative or non-proliferative stage and displays the results to the users. For image classification process, we have used Convolutional Neural Network and implemented VGG16 model. The images are processed followed by check constraints for over-fitting and under-fitting processes. Then image segmentation takes place, which helps in the accurate detection of the important characteristics such as exudates, micro-aneurysms, haemorrhages and optic disk. Clinically significant feature extraction for DR pathology detection and severity classification were performed using MATLAB R2019b with MATLAB Image Processing Toolbox. In addition, the graphical user interface was developed using the MATLAB GUI Toolbox. The accuracy of the software was measured by comparing the obtained results with the results of the diagnosis by an ophthalmologist. We evaluated the ability of the MATLAB system to predict the development of diabetic retinopathy in several ways. We evaluated discriminative power using the area under the receiver operating characteristic curve (AUC) and system calibration by plotting the observed event rate against the predicted event rate based on deciles of predicted risk. Furthermore, we evaluated the positive and negative predictive values of the system's predictions in all prediction percentiles. Confidence intervals for sensitivity, specificity, and positive predictive values and negative predictive values were calculated using the Pearson method based on the β distribution. Accurately capturing local information about the lesions present in images with DR such as micro-aneurysms, haemorrhages and hard exudates requires careful selection of features. We have used three different feature descriptors which capture the local information in different ways, and assessed their relative performance in our experiments Systematic screening for DR has been recognized as a cost-effective way to save health services resources. Automatic retinal image analysis is appearing as an important screening tool for early DR detection, which can reduce the workload related to manual grading as well as save diagnosis costs and time. Many research efforts in the last years have been done for developing automatic tools to help in the detection and evaluation of DR abnormal tissue.