Deep Multiple Instance Learning For Early Prediction Of Diseases From Diabetic Retinopathy In Retinal Images

Abstract

DR is a chronic disease which is the result of blood spillage in retinal vessels in diabetic patients. This can even cause vision loss at its adverse stage. The primary phase of the disease cannot be easily observed by the patients. Thus, an early prediction system is required to help out the patients to take necessary remedial therapy to avoid vision loss. In this work we proposed a deep MIL method to detect DR which learns features from the retinal images and also classifies it. Fuzzy-C-technique is deployed for segmentation and GLCM is deployed for extraction of features. ConvNet is deployed for classification. We have also proposed a methodology to predict the diseases that can attack the diabetic patients in future hinge on the extremity of retinopathy results. This method is very effective in predicting the early phase of fatal diseases which can be cured by proper diagnosis.

KEYWORDS: DR, Deep MIL, Fuzzy-C-Technique, Gray Level Co- occurrence Matrix, micro-aneurysm.