Recognizing Diabetic Retinopathy Using IoT Enabled Nonmydriatic Fundus Camera Images with help of Morphological Functions and Transductive Support Vector Machines

Recognizing Diabetic Retinopathy (DR) at the early stage is the major difficulty faced nowadays and it can be efficiently done using Fundus Images with the help of Morphological functions and Machine learning so that it enables early detection and timely treatment with IoT enabled Smartphone portable Fundus Camera. At early stage it does not show any symptoms as it progresses to proliferative level it causes blindness. Major indications of DR are appearance of microaneuryms, haemorrhages, and hard exudates. In this paper, an algorithm automatic detection of DR has been proposed including Morphological Structural Elements, Contrast Limited Adaptive Histogram Equalization (CLAHE), Median filter, Circular Hough Transform (CHT), thresholding and Wireless data transmission using Raspberry PI to Cloud Storage Services. Also in Machine learning, Support Vector Machines (SVM) classifier is used to classify fundus images to normal or presence of microaneurysm or haemorrhage or exudates. The image are captured through portable Nonmydriatic Fundus Camera connected to Raspberry Pi which uploads the image to Cloud Storage Services. The proposed algorithm has been tested for the images obtained from Cloud Storage database using MATLAB code. The sensitivity, specificity and accuracy of this approach are 93.33%, 90%, 95%.