Program Information
Reproducibility of Radiomics Features with Grow-Cut and Graph-Cut Semi-Automatic Tumor Segmentation in Hepatocellular Carcinoma
Q Qiu1,2*, J Duan1 , G Gong1 , D Li2 , Y Yin1,2 , (1) Shandong Cancer Hospital Affiliated to Shandong University, Jinan, Shandong province, China (2) Shandong Normal University, Jinan, Shandong Province, China
Presentations
SU-K-201-13 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: 201
Purpose: To evaluate the reproducibility of radiomics features extracted from CT images using Grow-Cut and Graph-Cut semi-automatic tumor segmentation algorithms in radiation therapy of hepatocellular carcinoma.
Methods: We selected arterial phase enhanced CT images of 15 primary hepatocellular carcinoma patients, the slice thickness was 3mm, CT interval was 3mm. The semi-automatic tumor segmentation of Grow-Cut and Graph-Cut were implemented in the publicly available software five times by two independent observers. For comparison, we also defined the tumor volume five times manually by five radiologists. Seventy-one radiomics features were extracted from the 3D tumor volume, including tumor intensity histogram, gray-level co-occurrence matrix(GLCM), neighbor intensity difference(NID), gray-level run length matrix(GLRLM) and shape. The intra-class correlation coefficient(ICC）was used to quantify reproducibility of radiomics features.
Results: We found that the radiomics features were extracted from the tumor volume which were defined by Grow-Cut and Graph-Cut algorithm showed higher reproducibility with the ICC (0.84±0.16, p=0.006) and (0.81±0.19, p=0.009), respectively. However, the ICC of radiomics features extracted from the manual delineation (0.78±0.13, p=0.008) was lower than semi-automatic methods. Compared with manual delineation, semi-automatic segmentation algorithms had better performance on the reproducibility of tumor intensity histogram features and texture features, but the reproducibility of shape features were worse.
Conclusion: More reproducible quantitative imaging features can be extracted from the tumor volumes which were segmented by semi-automatic algorithms. And, some differences of radiomics features existed between two semi-automatic segmentation algorithms. Therefore, it is very important to assess the reproducibility of radiomics features derived from tumor volume defined by different semi-automatic segmentation algorithms.
Contact Email: