Liver And Liver Tumor CT Image Segmentation And 3D Reconstruction For Tumor Treatment Electric Field Design

Liver cancer is one of the most common human cancers,and its fatality rate ranks among the highest.Currently,the widely used clinical treatment schemes,such as surgery,chemoradiotherapy,etc,all have significant side effects on the human body.The tumor treatment of electric field,a new cancer treatment method,is developing rapidly in the field of liver cancer treatment due to its characteristics of tumor growth inhibition,no obvious side effects on the human body,and remarkable effects in combination with drugs.Because of the differences between patients and tumors,it is necessary to conduct protocol design through personalized computer modeling analysis in advance in clinical practice.Since its introduction,computed tomography imaging is still the most commonly used method for liver cancer screening due to its high-cost performance and breakthrough in scanning technology.Based on cases of abdominal CT images for the liver and liver tumor segmentation and 3 d reconstructions can not only describe more details on anatomy,at the same time,by giving computer model dielectric characteristics of liver and liver tumor(EPS,including electrical conductivity and dielectric constant)in finite element modeling to TTFields quantitative analysis,improve the accuracy of the electric field distribution,the optimal design scheme.However,in the current clinical treatment planning or medical image segmentation research of liver cancer,the liver and tumor are generally segmented manually by doctors,which is very dependent on their professional level.At the same time,a set of CT often varies from hundreds to thousands of images,and the labeling process is tedious and energy-consuming.In addition,the current TTFields simulation studies on liver cancer assume that the EPS in the liver and liver tumor are uniform,without considering the heterogeneity in the tissue,and there are errors in the distribution of the electric field and the actual situation.This thesis mainly focuses on how to quickly and accurately segment liver and liver tumors based on CT images,and how to reflect the difference of tissue EPS in the segmentation result reconstruction model,to contribute to the design of personalized TTFields.Specific research contents are as follows:To investigate the range of common data types and tissue CT values in clinical and medical research of liver cancer,and to design segmentation schemes from the aspects of engineering application implementation and data set quality: When there are a large number of accurately labeled data sets,the method based on a convolutional neural network is adopted to solve the shortcomings of existing network segmentation of liver and liver tumors,such as liver boundary loss or tumor recognition failure,this thesis takes U-net network as the benchmark network,introduces the spatial attention mechanism,captures the edge pixels of the liver and breaks the local receptive field limitation.The experimental results on the public part of the LITS dataset show that the network with the attention mechanism can improve the performance of U-net,and the Non-Local based method has a more obvious effect in restoring the edge details.In the case of poor data set quality and limited computational power,aiming at the difficulty of poor contrast between liver and liver tumor in CT images and surrounding organs and tissues,a two-stage segmentation scheme was adopted.Firstly,the liver was segmtationed,and a preprocessing method based on the gray level conversion and edge-preserving filter was proposed.The canny edge prior information was added to the region-growing process to suppress oversegmentation,and the results were used as the initial level set to iterate the segmentation results.Liver tumors were segmented,and the results of one-stage liver segmentation were coarsely segmented using the improved OTSU3 threshold method,and the coarse segmentation results were put into the horizontal set for iterative refinement.The experimental results on 3Dircadb tumor-containing abdominal sequences show that the semi-automatic liver segmentation results with edge prior information can get the desired segmentation results,and the second-stage automatic tumor segmentation method has good performance in both speed and small tumor size.Finally,this thesis studies the close relationship between human tissue EPS,image CT value,and tissue density and the correlation mapping method,compare the effect of the reconstruction model based on VTK programming and software surface rendering and modularizes the segmentation and modeling process to enhance the applicability of this study.Specifically,the correlation between the CT value of liver tissue and EPS as well as tissue density and composition was firstly confirmed,and then the interval method was used to divide the liver CT interval into multiple segments according to the liver CT value survey data,the model of each part of liver and tumor was reconstructed by moving cube algorithm,reflecting the heterogeneity of EPS in the liver to a certain extent.