Abstract P6-02-19: Machine learning based analysis of CT radiomics for the simultaneous indeterminate pulmonary nodules of breast cancer

Abstract

Abstract PURPOSE: To investigate texture features of simultaneous indeterminate pulmonary nodules of breast cancer for predicting their potential metastasis. METHODS AND MATERIALS: 150 patients with simultaneous breast cancer diagnosed by biopsy and pulmonary nodules (diameter: 5-20mm) detected by preoperative CT were enrolled in this study. After surgery and breast cancer treatment, the patients were followed up for at least half a year or longer by CT to observe the changes of lung nodules, thereby inferring the potential of metastasis. We classify pulmonary nodules into two groups: the reduced or enlarged pulmonary nodules were defined as highly metastasis possibility (Group 1), and long-term stable pulmonary nodules were defined as low metastasis possibility (Group 2). In addition, pathologic proven primary lung cancer in this study (Group 3) was compared with Group 1. Therefore, we carried out a comparative analysis of the texture features between the groups, and additional statistical were used three regression testing to extract texture features. Finally, we construct a machine learning classifier and calculate the accuracy of cross-validation. RESULTS: We collected 106 features by the texture analysis(TA). There are 18 features with significant differences between Group 1 and the Group 2(p<0.05), and 76 features with significant differences in the Group 1 and Group 3 (p<0.05). We tried to find key features related to pathology in 106 features using three methods: lasso regression, ridge regression and forward stepwise regression. The accuracy in different regressions respectively is 94.5%,94.5%,89.7% using KNN between Group 1 and Group 2. The accuracy in different regressions respectively is 96.2%(KNN),96.2%(Tree),92.3%(Linear Discriminant)in the Group 1 and Group 3. CONCLUDES: The identified radiomics features have the potential to be used as a biomarker for metastasis prediction of simultaneous indeterminate pulmonary nodules in breast cancer patients, and it may contribute to preoperative treatment and postoperative follow-up planning. Citation Format: Xiao Q, Gu Y, Wu J, Wang Z, Huang Y. Machine learning based analysis of CT radiomics for the simultaneous indeterminate pulmonary nodules of breast cancer [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P6-02-19.