A convolutional neural network with pixel-wise sparse graph reasoning for COVID-19 lesion segmentation in CT images.

The COVID-19 pandemic has extremely threatened human health, and automated algorithms are needed to segment infected regions in the lung using computed tomography (CT). Although several deep convolutional neural networks (DCNNs) have proposed for this purpose, their performance on this task is suppressed due to the limited local receptive field and deficient global reasoning ability. To address these issues, we propose a segmentation network with a novel pixel-wise sparse graph reasoning (PSGR) module for the segmentation of COVID-19 infected regions in CT images. The PSGR module, which is inserted between the encoder and decoder of the network, can improve the modeling of global contextual information. In the PSGR module, a graph is first constructed by projecting each pixel on a node based on the features produced by the encoder. Then, we convert the graph into a sparsely-connected one by keeping K strongest connections to each uncertainly segmented pixel. Finally, the global reasoning is performed on the sparsely-connected graph. Our segmentation network was evaluated on three publicly available datasets and compared with a variety of widely-used segmentation models. Our results demonstrate that (1) the proposed PSGR module can capture the long-range dependencies effectively and (2) the segmentation model equipped with this PSGR module can accurately segment COVID-19 infected regions in CT images and outperform all other competing models.

3D bi-directional transformer U-Net for medical image segmentation.

As one of the popular deep learning methods, deep convolutional neural networks (DCNNs) have been widely adopted in segmentation tasks and have received positive feedback. However, in segmentation tasks, DCNN-based frameworks are known for their incompetence in dealing with global relations within imaging features. Although several techniques have been proposed to enhance the global reasoning of DCNN, these models are either not able to gain satisfying performances compared with traditional fully-convolutional structures or not capable of utilizing the basic advantages of CNN-based networks (namely the ability of local reasoning). In this study, compared with current attempts to combine FCNs and global reasoning methods, we fully extracted the ability of self-attention by designing a novel attention mechanism for 3D computation and proposed a new segmentation framework (named 3DTU) for three-dimensional medical image segmentation tasks. This new framework processes images in an end-to-end manner and executes 3D computation on both the encoder side (which contains a 3D transformer) and the decoder side (which is based on a 3D DCNN). We tested our framework on two independent datasets that consist of 3D MRI and CT images. Experimental results clearly demonstrate that our method outperforms several state-of-the-art segmentation methods in various metrics.