Fundus problems are the most typical causes of blindness in people globally. The ophthalmic disease is notewor�thy, since it has features that are irreversible and might result in long-term blindness. The majority of the identification models in use exclusively concentrate on one particular ocular disease. Therefore, My goal was to create a model for automatically classifying many ocular diseases using fundus photos as input and reporting the disease’s name if it is present.There are currently a number of models available, but the benefit is that I am using multi�labeled picture datasets as opposed to binary-labeled data. The multilabelled data covers disorders such as age-related macular degeneration, cataract, diabetes, glucoma, pathological myopia, and hypertension.The multilabelled data covers disorders such as age-related macular degeneration (AMD), diabetes, glucoma, pathological myopia, and hypertension.
The dataset utilised in this investigation is Ocular Disease Intelligent Recognition. One of the most extensive public resources on Kaggle for identifying eye illnesses is this dataset. Classifications of ocular diseases are used to group the fundus photos in this collection. Myopia (M), hypertension (H), diabetes (D), cataract (C), glaucoma (G), and age-related macular degeneration (A) are among the conditions. Normal is the seventh one. The 5000 colour fundus images in this dataset are split into training and testing groups. All of the photographs for this project were scaled to 224x224.
MATLAB2022a, Pre-trained Model- ResNet-50, Multiclass SVM, Deep Learning
I have taken a multi-class disease dataset. Feature extraction is done by Resnet-50 and VGG-16 while Classification is done by using Multi-class SVM.
- Collecting Dataset from kaggle
- Using equal number of images from unbalanced data categories labels
- Train and Test Split
- Model Building using ResNet-50
- Testing model
- Prediction using multiclass SVM
I have provided a novel approach to categorise retinal fundus pictures into categories: myopia, hy�pertension, diabetes, cataract, glaucoma and age-related macular degeneration using deep learning models (ResNet�50) and multiclass SVM. To solve the problems of CNN model training on a limited dataset, the pre-trained models are initially employed for automated feature extraction. Then SVM, which has previously demonstrated excellent generalizability, is employed for disease classification. The testset will be used to extract picture characteristics using the same process. The classifier may then be given the test features in order to evaluate how accurate the trained classifier is. Hence, the Resnet-50 model for feature extraction and then using multi-class SVM as a classifier to classify images gives accuracy of 74%.