Purpose:

To investigate whether radiomics based on contrast-enhanced MRI can predict pathological complete response(pCR) and disease-free survival(DFS) of locally advanced TNBC undergoing neoadjuvant chemotherapy(NAC).

Materials and Methods:

A total of 113 patients who received NAC from January 2010 to May 2017 were retrospectively analyzed, and then follow up to August 2018.These patients were diagnosed with triple-negative breast cancer through needle biopsy. Pathological complete response was defined as ypT0/is and ypN0.Dynamic imaging was performed at 1.5T MR. ROIs were drawn on T1W enhancement images on the whole volume of tumor with 3D slicer, including necrotic or cystic parts. Radiomics signature was used to predict pathological complete response after NAC. Univariate and multivariate Cox proportional hazards model and Kaplan-Meier analysis were used to determine the association of the radiomics signature, MRI findings, and clinicopathological variables with DFS. The Radiomics signature includes 1177 radiomics features. The 15 shape-based features were calculated on the first postcontrast DCE images. The 83 texture features (including 18 first order features, 23 gray level co-occurrence matrix (GLCM) features, 13 gray level run length matrix (GLRLM) features, 13 gray level size zone matrix (GLSZM) features, 5 neighboring gray tone difference matrix (NGTDM) features, and 11 gray level dependence matrix (GLDM) features) were calculated on these six series image sets to yield 332 features. To characterize the textural changes on DCE images over time serials, we measured ten new sequential features for each texture feature described in group b (supplementary Table S2). The first six features, including mean, variance, kurtosis, skewness, energy, and entropy, were extracted from each individual subject. The other four features, including Kendall-tau-b, conservation, stability, and dispersion, were calculated from the interactive information between the current subject and the remainder of the subjects. Therefore, a total of 830 DCEsequential features were extracted from 83 texture features.

Results:

All the machine learning models were conducted using the 5-fold cross validation, whereby 20% of the data were used to test the model created by the other 80% of the data. 47 of 113 (41.6%) patients show pathologic complete response (pCR) and 66 of 113 (58.4%) show nonpathologic CR (npCR). The median follow-up time is 36 months. 28 of 113(24.7%) occurred recurrence and metastasis. The model for predicting pCR shows that sensitivity and specificity are 77.0% and 88.0%, respectively; The accuracy rate yields 84.1%( area under ROC curve: 0.87). The model for predicting DFS shows that sensitivity and specificity are 71.0% and 94.0%, respectively; the accuracy rate yields 88.5%( area under ROC curve: 0.87).

Conclusion:

The radiomics signature based on DCE-MR images is an independent biomarker for the early prediction of pCR and DFS in patients with TNBC to NAC pretreatment of chemotherapy.

Key words:

Radiomics ; TNBC; neoadjuvant chemotherapy; Machine learning

Acknowledgements

We recognize Dr. Chao You, Dr. Tong Tong and Dr. Bin Wu for their discussions of the study design and research results. This work was supported by the National Natural Science Foundation of China (no. 61731008). This project has also been funded by Shanghai Municipal Science and Technology Major Project (no. 2017SHZDZX01) and Shanghai Natural Science Foundation (no. 17ZR1401600).