Automatic Liver Segmentation And Three-Dimensional Imaging Based On CT Image

The liver is a very important organ in the human body,and in recent years,the proportion of diseases caused by the liver is increasing.Currently,doctors mainly observe the liver through images obtained by various medical devices.The limited information provided by two-dimensional medical images poses certain difficulties for doctors’ diagnosis.A three-dimensional liver model can allow doctors to observe the condition more intuitively.Most of the existing three-dimensional reconstruction methods for medical images are based on the threshold differences of different organs in the image to construct three-dimensional models,and the results of three-dimensional model construction are prone to interference from other organs and tissues.In response to the above issues,the main research work done in this thesis is as follows:(1)A three-dimensional liver reconstruction method based on segmented images is proposed.The Marching Cubes algorithm constructs a three-dimensional model by establishing an isosurface.In CT images,some organs have close distances and small differences in CT values,which are difficult to distinguish based on threshold values,interfering with the construction of three-dimensional models.In response to the above issues,this paper proposes a three-dimensional liver model reconstruction method based on segmented images.First,the liver is segmented,and then three-dimensional reconstruction is performed using the segmented image.This solves the problem of inter-organ threshold proximity interference in three-dimensional model construction.(2)A FANU-Net network based on human-computer collaboration is proposed.Traditional segmentation methods such as threshold and edge detection are not suitable for medical image segmentation.The segmentation effect of the U-Net under deep learning has been improved.However,due to irregular liver changes and no obvious features in shape,the segmentation results are still not ideal.In this paper,the Fully Attentional Networks module and human-computer interaction are added to the U-Net to improve the segmentation effect of the model on the liver by enhancing the feature extraction ability of the model and manually modifying the training data.(3)A sequential based Siam-FANU-Net network is proposed.Most of the existing segmentation methods based on deep learning focus on feature extraction of the target,but the liver shape in different CT images of different people or even different CT images of the same person is different,making it difficult to extract features.The CT image is a continuous slice image in which the liver has continuity despite changes,so its continuity can be used to assist segmentation and improve the segmentation results of the network.We replaced the segmentation module of the Siam Mask with the U-Net,which improved the network’s segmentation ability,and added the Fully Attentional Networks module to improve the network’s feature extraction ability.The segmentation effect of the model was significantly improved.