iamraufu / decipherable-classification-of-glaucoma-using-deep-neural-network-leveraging-xai

Decipherable Classification of Glaucoma usingDeep Neural Network Leveraging XAI

Decipherable-Classification-of-Glaucoma-using-Deep-Neural-Network-Leveraging-XAI

Glaucoma is the second driving reason for partial or complete blindness among all the visual deficiencies which mainly occurs because of excessive pressure in the eye due to anxiety or depression which damages the optic nerve and creates complications in vision. In this research, we used the Glaucoma Dataset in our algorithm to predict outcomes related to Glaucoma, suspicious glaucoma, and non-glaucoma. The main goal of the author of this research was to develop an automated deep learning neural network architecture for early detection of Glaucoma disease.For the classification of glaucoma three Black Box models have been used in the paper, such as Fully Connected Neural Network (FCNNs), Support Vector Machine (SVM), and Conventional Neural Network (CNN). This Black Box model has been described through Explainable Artificial Intelligence (XAI) to achieve the ultimate goal of our research. However, to serve our purpose we have used VGG-16, VGG-19, DenseNet121, InceptionV3 and ResNet50 models for our study. To begin, we pre-processed the images and grouped them into three sets: training, testing, and validation. Afterwards, DCNN models have been initialized with the pre-existing models trained on the imagenet dataset. Conclusively, the training and evaluating of all the DCNN has been done. The validation accuracy of our models we got are as follows: InceptionV3 we got 86.4% accuracy, in DenseNet121 we got 86.8% accuracy, in ResNet50 we got 94.7% accuracy, in VGG-19 we got 93.3% accuracy and lastly in VGG-16 we got 88.6% accuracy. As follows, after 50 epochs, RestNet50 got the highest score among the other models with a validation accuracy of 94.7%. Afterwards we compared all models' accuracy and loss graph together, where we can see that VGG-19 and ResNet50 were the Good-Fit than the other models. As a result, our research achieved outstanding classification accuracy in a short period of time. However, it seems to be vital to understand that a human can rely on black-box level Deep Learning models to make decisions. Throughout this work, a hybrid approach combining image processing with deep learning has been used with the support of XAI to assure reliable glaucoma detection at an early stage.