Deep learning

Image segmentation

Pattern recognition (psychology)

Fundus (uterus)

Diabetes mellitus

Biology

Ophthalmology

Endocrinology

Stage (stratigraphy)

Paleontology

Artificial intelligence

Diabetic retinopathy

Retinopathy

Computer science

Computer vision

Medicine

Convolutional neural network

Segmentation

One of the commonly found medical conditions in people with type 1 or 2 diabetes is Diabetic Retinopathy (DR). It is mostly asymptomatic in its early stages and can lead to both central and side vision loss if left untreated. The five stages of this disease include normal (no DR), mild stage non-proliferative DR, moderate stage non-proliferative DR, severe stage non-proliferative DR, and proliferative DR. This condition is generally diagnosed by a highly trained ophthalmologist by examining the retinal images of the patients. But the medical image analysis by humans can be prone to perceptual and cognitive errors. On the other hand, deep learning has proven to be successful in medical image analysis and medical image segmentation in the last few decades. There are already many deep learning-based computer vision techniques that are available for DR detection using fundus retinal images. But most of them don't classify all the stages of DR accurately. The proposed method uses an ensemble model of five deep Convolutional Neural Networks (CNN) trained on the EyePACS dataset images from Kaggle. Each channel of the image (Red, Green, Blue) is fed separately to all the models and the effect of individual channel on the result is analyzed. The proposed method performs better in classifying all the stages than most of the existing methods that use the same dataset.

Diabetic Retinopathy Detection using ensemble deep Learning and Individual Channel Training