Detecting Anemia from retinal images using Deep Learning

Abstract

Anemia is the condition which formed when there is not enough healthy red blood cells present in body. This ultimately result in the lack of reduced oxygen flow in the body. This causes the fatigue, dizziness, shortness of breath. The main problem with this is it’s very common and mostly undetected specially in the Indian population. In general anemia can be detected by the blood test and checking the hemoglobin level with standard WHO standards of men and women to decide anemia or non anemia which is an invasive method. Through some studies, it is been observed that their is an non-invasive way to detect anemia as well, which is by using medical fundus images. It can be used to detect this disease via the Deep learning network. In one of that research [4] medical fundus images with the centered macula region is being used with the huge dataset but in the end their were some evidences in paper that model try to focus on the optic disc region much more and based on this prediction do get affected. We have here the dataset of the medical fundus images provided by the eye hospital and we are picking out images each with the optic disc present, in which we are going to be using for the classification of the anemic and non anemic disease based on the deep learning model. In the end of the experiment we have seen that with AUC 0.6649 in Densenet121, 0.6223 in Resnet50 and 0.6718 in MobilnetV2 we are able to identify the disease.