Efficient diabetic retinopathy classification grading using GAN based EM and PCA learning framework

Diabetic retinopathy (DR), a type of eye disease, is a danger for diabetics. Manual labour, which is prone to inaccuracy and time consuming, makes dealing with this illness considerably more difficult. Normally computer-assisted diagnosis has appeared as a promising tool for the early identification and severity grading of DR. As technologies are revolutionizing day by day, in which the most advance technology deep learning's algorithm gives a tremendous support for healthcare fields. This article proposes an efficient classification of DR models for categories the DR into different grades and to identify the severity. There various prediction techniques employed in DR detection. Radial Basics Network, Multilayer Perceptron and Recurrent Neural Network are binary classifiers employed for DR classification. Further the Bag of Visual Words and Convolutional Neural Networks implements for the stages of 3. The performance shows that Convolutional Neural Network perform superior over other methods and attains 98.3%. It is of great significance to apply deep-learning techniques for DR recognition. However, deep-learning algorithms often depend on large amounts of labeled data, which is expensive and time-consuming to obtain in the medical imaging area. In addition, the DR features are inconspicuous and spread out over high-resolution fundus images. Therefore, it is a big challenge to learn the distribution of such DR features. To overcome this, This research work proposes a multichannel-based generative adversarial network (M-GAN) for data augmentation as well as classification to grade DR The usefulness and effectiveness of GAN for classification of fundus images are explored for the first time.Obtaining medical data is also a tedious and challenging one because it is quite expensive and confidential, to overcome this proposed model is acts data augmentation model, moreover the features in the input data’s are reduced by Dimensionality reduction Module (DRM) based on Principal Component Analysis. When compared to previous classification models, experimental results on fundus images using the MESSIDOR database reveal that the GAN-based EM model obtained good processing performances with 99.2% accuracy in the suggested classification task. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.