Multi-Sections Fusion Method Based on VGGU-Net for Liver CT Image Segmentation

Automatic and accurate segmentation of organs in abdominal CT images is helpful for disease diagnosis and surgical planning. The liver organ segmentation method based on two-dimensional neural network technology has achieved high accuracy, but its segmentation accuracy is still limited due to the lack of three-dimensional feature extraction for the liver. In this paper, we propose a new two-dimensional liver segmentation method. This method extracts two-dimensional liver features from three different cross-sections of abdominal CT images and fuses them in a certain way to improve liver segmentation accuracy. In order to better extract two-dimensional features of the liver, a new U-shaped neural network architecture is proposed here. This architecture is a combination of VGG19 and U-Net. To verify the effectiveness of this method, we tested it on a publicly available dataset and compared it with other methods. The experimental results show that our proposed method has the highest accuracy in liver segmentation, with an average Dice coefficient of 95.81%.