OXENDONET: A dilated convolutional neural networks for endoscopic artefact segmentation

Medical image segmentation plays a key role in many generic applications such as population analysis and, more accessibly, can be made into a crucial tool in diagnosis and treatment planning. Its output can vary from extracting practical clinical information such as pathologies (detection of cancer), to measuring anatomical structures (kidney volume, cartilage thickness, bone angles). Many prior approaches to this problem are based on one of two main architectures: a fully convolutional network or a U-Net-based architecture. These methods rely on multiple pooling and striding layers to increase the receptive field size of neurons. Since we are tackling a segmentation task, the way pooling layers are used reduce the feature map size and lead to the loss of important spatial information. In this paper, we propose a novel neural network, which we call OxEndoNet. Our network uses the pyramid dilated module (PDM) consisting of multiple dilated convolutions stacked in parallel. The PDM module eliminates the need of striding layers and has a very large receptive field which maintains spatial resolution. We combine several pyramid dilated modules to form our final OxEndoNet network. The proposed network is able to capture small and complex variations in the challenging problem of Endoscopy Artefact Detection and Segmentation where objects vary largely in scale and size