Classification of Diabetic Retinopathy Using Deep Learning Techniques

Diabetes is a disease unfolding to be a big threat to humanity, which even after such scientific and
medical advancement remains incurable. Its only remedy is early detection and precautionary measure
to scale back its effects to a minimum. Diabetic retinopathy is a diabetes complication that affects our
eyes. in consequence of damage to the blood vessels of the light-sensitive tissue at the back of the eye
(retina).
Initially, diabetic retinopathy may cause no symptoms or only mild vision problems. Eventually, it can
cause blindness. The condition can evolve in anyone who has type 1 or type 2 diabetes. The longer
you've diabetes and thus the less controlled your blood sugar is, the more likely you're to develop this
eye complication. This disease affects every age group of people. In 2015, International Diabetes
Federation performed a survey and mentioned that nearly 410 million people are suffering from this
disease, globally. In 2015, Diabetes has caused more than 5 million deaths. India is a dwelling place for
approximately 70 million people with diabetes, and this epidemic is estimated to increase to 130 million
by 2045. India also has the greatest number of people with blindness worldwide. Diabetic Retinopathy
(DR) is the most common complication of diabetes, and approximately 2.6% of global blindness is
caused by diabetes. In absolute terms, approximately 3–4.5 million people in India are estimated to suffer
from vision-threatening diabetic retinopathy (VTDR). The treatment options for VTDR need costly
devices and medications and therefore the disease requires regular follow-up from diagnosis to the endof-
life. Given that 70% of the population of India relies on out-of-pocket expenses for their healthcare,
one person with VTDR in a household is sufficient to drive a family to below the poverty line. Therefore,
all measures should be initiated urgently to stop people with diabetes to enter the vicious circle of
diabetes, blindness, and poverty. So, our proposed work will classify the stages of Diabetic Retinopathy
with an approximate precision of 97.8% which is an improvement from the previous model with a
margin of around 10%. We have used pre-trained models such as VGG16, ResNet50, DenseNet201, and
many more, but the most accurate result came when we used the ensemble learning process with these
models.