Multi-label Classification of Retinal Diseases using Hybrid Vision Transformer

Retina is the rear side of eye which behaves like a screen for image formation. Retina converts the light energy into the signals for passing these signals further to brain. It is the irreplaceable part of human eye so taking care of retina is the essential task. There are various major retinal diseases like optic disc cupping, age-related macular degeneration, glaucoma, optic disc edema, diabetic retinopathy, etc. which should be detected at earlier stage. Artificial intelligence and machine learning can play a vital role in early detection of retinal diseases and now a days help in improving the quality of care. For detecting these various retinal diseases simultaneously from fundus images, multilabel classification can be used. Multilabel classification predicts each label independently. Firstly, convolutional neural network is used over VIETAI datasets. Transfer learning is also used for training the model in which some pretrained models like DenseNet121, VGG16, ResNet50 are used. These pre-trained models help in enhancing the accuracy. For more improvement in results, vision transformer is used. Then, we used a hybrid model which is combination of convolutional neural network and vision transformer named as hybrid vision transformer. Hybrid vision transformer outperforms of all used algorithms for this multi-label classification problem.