Pyramidal deep neural network for classification of retinal OCT images

Retinal optical coherence tomography (OCT) images are widely used to diagnose and grade macular diseases, such as age-related macular degeneration (AMD). However, manual interpretation of OCT images is time-consuming and subjective. Therefore, automated and accurate classification of OCT images is essential for assisting ophthalmologists in clinical decision-making. This paper proposes a pyramidal deep neural network that can diagnose normal and two types of AMD (dry and wet) in OCT images. Our network leverages features from different scales of a pre-trained convolutional neural network (CNN) and integrates them with two advanced versions of feature pyramid networks: bidirectional feature pyramid network (BiFPN) and path aggregation network (PANet). We evaluate our network on the NEH dataset and compare it with its predecessor. Our results show that our BiFPN-VGG16 and PAN-VGG16 models achieve accuracies of 94.8% and 95.0%, respectively, which are 2.8 to 3% higher than the previous models. Our approach demonstrates the potential of multi-scale feature networks for OCT image classification and can serve as an auxiliary diagnostic tool for ophthalmologists.