Detection of Diabetic Retinopathy in Fundus Images Using SVM
Atharva Vyas1, Chandan Suthar2, Ishan Sinnarkar3, Varad Vaidya4, Varad Wagh5, Noshir Tarapore6
1LY B.Tech, Computer Engineering, Science & Technology, Vishwakarma University, Pune, India - 411048
2LY B.Tech, Computer Engineering, Science & Technology, Vishwakarma University, Pune, India - 411048
3LY B.Tech, Computer Engineering, Science & Technology, Vishwakarma University, Pune, India - 411048
4LY B.Tech, Computer Engineering, Science & Technology, Vishwakarma University, Pune, India - 411048
5LY B.Tech, Computer Engineering, Science & Technology, Vishwakarma University, Pune, India - 411048
6Assistant Professor, Dept. of Computer Engineering, Vishwakarma University, Pune, India - 411048
Abstract - One of the main causes of blindness in the industrialised world, diabetic retinopathy is a consequence of diabetes brought on by alterations in the retina's blood vessels. Patients who have diabetic retinopathy are protected from losing their vision by early identification. The automatic detection of diabetic retinopathy using the support vector machine algorithm is the sole goal of this project. The major objective is to automatically assess any fundus image for non-proliferative diabetic retinopathy. The fundus image that is used as the project's input is sent to the preprocessing stage for noise removal. It separates blood vessels, microaneurysms, and hard exudates during the image preprocessing stage. The feature extraction procedure is applied to the preprocessed image. The SVM classifier receives the retrieved image and uses it to determine each fundus image's retinopathy grade. Additionally, the classifier will determine whether or not the image has diabetic retinopathy. In order to help individuals identify diabetic retinopathy early on, this research suggests a computer-assisted diagnosis based on the digital processing of fundus images. A library of 400 fundus pictures that have been classified using a 4-grade scale for non-proliferative diabetic retinopathy has been used to evaluate this concept. As a result, we were able to forecast data with a 90% accuracy. It has also been assessed how resilient the algorithm is to changes in its parameters.
