Impact Of Different Tumor Segmentation Methods On Extraction Of CT Radiomic Features In Hepatocellular Carcinoma

Hepatocellular carcinoma(HCC)is one of the most prevalent cancers in the world and has a poor prognosis.With the development of personalized medicine and computer technology,Radiomics has become an emerging field in medical imaging research and play an essential role in cancer staging,treatment planning and outcome monitoring.The Radiomics is a technique based on the quantitative analysis of image features,and its image features must have three characteristics: reproducibility,non-redundancy and informative.Only the image feature with these three characteristics can be used for clinical study.However,there are many factors in the workflow of Radiomics that affect the reproducibility of radiomic feature,and the tumor segmentation is an important factor.Since liver tumor usually with indistinct boundary,manual contour often introduce significant inter-observer variation in the segmentations and the impact on radiomic features should be evaluated.In this study,we evaluate the reproducibility and stability of quantitative imaging features derived from tumor volume segmented by using Graph Cut and Grow Cut interactive methods from arterial CT images of HCC patients.Meanwhile,the dimensionality reduction of reproducible features was done by using hierarchical clustering method to reduce the redundancy of features.The results show that the reproducibility and non-redundancy of the radiomic features rely greatly on the tumor segmentation in HCC CT images.Our study show that semi-automatic segmentation is likely to increase the reproducibility of imaging markers and hierarchical clustering can provide robust radiomic feature clusters and reduce feature redundancy.Furthermore,to guarantee the segmentation precision and maximally eliminate segmentation effects,a proper semi-automatic algorithm should be considered for various tumors with different imaging modalities.Main contents in this paper are as follows:1.We introduce the development status of HCC radiomic in the first chapter.We also introduce the development process and existing problems of Radiomics and the purpose of this study.2.We introduce the semiautomatic segmentation method and the characteristics of three-phase enhanced CT images of liver tumor.Meanwhile,the results of manual and two semiautomatic tumor segmentations were compared.3.We summarized the related software in Radiomics and the extracted features in this study were introduced.The mathematical description of these features was given.4.Combined with intra-class correlation coefficient(ICC),we analyzed the reproducibility of radiomic features.Then,hierarchical clustering was used to reduce the redundancy of these reproducible features.Finally,the stable,reproducible and non-redundant features were acquired.5.Discuss the results of this study,explain the reasons for the results,and summarize the content of this study.