A Liver Tumor Image Segmentation Method Based on Multi-Scale Feature Fusion and Reconstruction Convolution

Aiming at the problem of insufficient expression ability of liver tumor image features and limited global contextual information transmission, an improved U-Net liver tumor image segmentation method is proposed. Firstly, a low-rank reconstruction convolution is designed to optimize the large number of parameter problems caused by traditional convolution operations, and is used to construct a convolution kernel reconstruction module that uses residual structure to improve the encoder decoder, so that the encoder retains more detailed information and the decoder recovers information more effectively, thereby enhancing the expression ability of liver tumor image features. Then, to enrich the transmission of global contextual information, a three-branch spatial pyramid pooling module is designed to optimize the bottleneck structure of information transmission and to break the limitation of a single path. Secondly, a multi-scale feature fusion module is designed to optimize the reuse mechanism of encoder information, enhance the modeling ability of the model for global contextual information, and improve its efficiency in extracting liver tumor image features in different scales. Finally, the performance of the proposed method is tested on LiTS2017 and 3DIRCADb datasets. Experimental results show that the method achieves a Dice coefficient and an IoU value of 97.56% and 95.25% in the liver image segmentation task on LiTS2017 dataset, and 89.71% and 81.58% in the liver tumor image segmentation task. Moreover, the Dice coefficient and IoU value in the liver image segmentation task on 3DIRCADb dataset respectively reach 97.63% and 95.39%, while respectively reach 89.62% and 81.63% in the liver tumor image segmentation task.