In this project we aim to diagnoise diabetic retinopathy using Gray World Color Constancy Algorithm and Stacked generalisation of three different Convolutional Neural Networks and draw Comparitive analysis with pretrained Transfer Learning Models such as resNet50 and VGG-16
Manuscript abstract:
Diabetic retinopathy (DR) is a diabetes complication that affects the eye and can cause damage from mild vision problems to complete blindness. It has been seen that the eye fundus images shows various kinds of colour aberrations and irrelevant illuminations which degrade the diagnostic analysis and may hinder with the results. Thus, in this paper we present a methodology to eliminate these unnecessary reflectance properties of the images using a novel image processing scheme and a stacked deep learning technique for the diagnosis. For the luminosity normalisation of the image, the Gray world colour constancy algorithm is implemented which does image desaturation to a great extent and thus improves image quality. The effectiveness of the proposed enhancement technique is evaluated on the basis of Peak signal to noise ratio (PSNR) and mean squared error (MSE) of the normalised image. To automate the detection process using deep learning techniques, we present a stacked generalisation of the multilayer Convolution neural networks (CNN). Three custom CNN models are fed into a single meta-learner for combining the most optimum weights to obtain superior metrics of evaluation and thus leading to promising prediction results. Extensive experimental results reveal that the proposed methodology of illumination normalisation greatly facilitated the deep learning model and yields better results than various state-of-art techniques.
Dataset Acquired from : http://www.eyepacs.com/data-analysis
