ECNet: An evolutionary convolutional network for automated glaucoma detection using fundus images

Glaucoma is an ocular disorder that affects the optic nerve and ultimately leads to partial or complete vision loss. Hence, there is a strong need for early screening of glaucoma. Earlier diagnosis schemes mostly rely on handcrafted feature engineering. On the other hand, the non-handcrafted feature extraction methods are generally designed with the support of gradient-based algorithms that suffer from critical problems like overfitting and demand for a larger set of samples for effective training. To mitigate these issues, in this paper, we propose a novel non-handcrafted feature extraction method termed as evolutionary convolutional network (ECNet) for automated detection of glaucoma from fundus images. The proposed method includes various layers such as convolutional, compression, rectified linear unit (ReLU), and summation layer which facilitate the extraction of discriminative features. An evolutionary algorithm called real-coded genetic algorithm (RCGA) is employed to optimize the weights at different layers. The ECNet is trained using a criteria that maximizes the inter-class distance and minimizes the intra-class variance of different classes. The final feature vectors are then subjected to a set of classifiers such as K-nearest neighbor (KNN), backpropagation neural network (BPNN), support vector machine (SVM), extreme learning machine (ELM), and kernel ELM (K-ELM) to select optimum performing model. The experimental results on a dataset of 1426 fundus images (589 normal and 837 glaucoma) demonstrate that the ECNet model with SVM yielded the highest accuracy of 97.20% compared to state-of-the-art techniques. The proposed model can aid ophthalmologists to validate their screening.