A Lightweight 3D Segmentation Network for Abdominal Liver in CT Image

Abnormal liver function is linked to a variety of disorders. Precise and quick automatic liver segmentation can help clinicians make better diagnosis and treatment decisions. With the development of computer vision and deep learning approaches, there are more solutions for biomedical image segmentation tasks. In recent years, the U-Net architecture is by far the most widely used backbone architecture for biomedical image segmentation. Deep convolutional neural networks-based semantic segmentation has achieved sufficient accuracy. However, the scale of high-precision networks is growing, requiring an increasing amount of storage and computational resources. Furthermore, the deep neural network's operating time is lengthy, making it difficult to satisfy practical needs. As a result, the lightweight convolutional neural network design is used to the semantic segmentation task. As a consequence, in this article, a lightweight convolutional neural network is proposed to solve the aforementioned problems in the task of biomedical image segmentation. 3D U-Net is used as the backbone architecture and a modification of the Ghost module from GhostNet is introduced to boost up the effectiveness and the learning efficiency. The experimental results demonstrate that the proposed network improved the segmentation performance with fewer network parameters and requiring less floating-point computation.