Automated detection of age-related macular degeneration using a pre-trained deep-learning scheme

An eye disease affects the entire sensory operation, and an unrecognised and untreated eye disease may lead to loss of vision. The proposed work aims to develop an automated age-related macular degeneration (AMD) detection system using a Deep-Learning (DL) scheme with serially concatenated deep and handcrafted features. The research includes the following phases: initial data processing, deep-features extraction with VGG16, handcrafted feature extraction, optimal feature selection using Mayfly-Algorithm, serial features concatenation, and binary classification and validation. In this work, the handcrafted features, such as local binary pattern (LBP), pyramid histogram of oriented gradients (PHOG), and discrete wavelet transform (DWT), are extracted from the test images and concatenated with the deep-features of VGG16. The performance of the developed system is separately tested using fundus retinal images (FRI) and optical coherence tomography (OCT) images. A binary classification with a fivefold cross validation is employed and the best outcome of this trial is chosen as the result. The performance of VGG16 is then compared with that of VGG19, ResNet50, and AlexNet. The experimental outcome of this research confirms that the proposal of VGG16 with concatenated features achieves AMD detection accuracies of 97.08 and 97.50 for FRI and OCT images, respectively.