|International Journal of Advances in Electronics and Computer Science-IJAECS
|Volume-8,Issue-10 ( Oct, 2021 )
|Retinal Image Synthesis for Diabetic Retinopathy Assessment using DCGAN and VAE models
|Y.Sravani Devi, S.Phani Kumar
|Research Scholar, GITAM Deemed to be University HOD-CSE, GITAM Deemed to be University
|Amongst the most significant considerations in classification of the image is the data amounts particularly in the medical images. Although the main challenge in the healthcare sector is „attaining the datasets. In this, we display the images of the synthesized retinal fundus by preparing a VAE i.e., Variational Autoencoder & another model known as the DCGAN, adversative model on almost 3662 images of retina which have been captured from a dataset known as the APTOS- Blindness dataset. The finding of this method is in creating the images of retina without the usage of vessel segmentation that is previously used. This enables the system to become independent. The models which are acquired are the synthesizers of the image that are proficient in producing resized images of retina of any amount from a basic regular distribution. Moreover, a lot of images than this have been utilized for the purpose of training than any other task in literature. The assessment or appraisal of a synthetic image is done by giving an output to a CNN model & the average squared error was counted between the average 2-Dimensional hologram of images that were real and synthetic as well. Later, by analyzing the latent space and average loss of the images. The achieved outcomes out of the analysis inferred that the general images have less extent of loss in DCGAN as opposed to Variational Auto Encoders. Keywords - Diabetic Retinopathy, Data Augmentation, Generative Adversarial Network, DCGAN, Variational Auto Encoder.