4878

MR-based machine learning radiomics can predict tumor heterogeneity and pathologic response after neoadjuvant therapy in HER2 breast cancer
Almir Bitencourt1,2, Peter Gibbs3, Carolina Rossi Saccarelli3, Isaac Daimiel3, Roberto Lo Gullo3, Michael Fox3, Sunitha Thakur3, Katja Pinker3, Elizabeth A Morris3, Monica Morrow4, and Maxime Jochelson3
1Breast Radiology, MSKCC, New York, NY, United States, 2Department of Imaging, A.C. Camargo Cancer Center, Sao Paulo, Brazil, 3MSKCC, New York, NY, United States, 4Breast Surgery, MSKCC, New York, NY, United States

Synopsis

In this study, we used magnetic resonance (MR)-based clinical and radiomic features to assess tumor heterogeneity in 311 HER2 overexpressing breast cancer patients receiving neoadjuvant chemotherapy (NAC), and correlated these findings with tumor heterogeneity and pathologic response. Tumor heterogeneity was evaluated based on the HER2 expression (IHC vs. FISH) . Pathologic complete response (pCR) was defined as no residual invasive carcinoma in the breast or axillary lymph nodes (ypT0/isN0). Radiomics analysis and machine learning with MRI were able to assess tumor heterogeneity and predict pCR after neoadjuvant chemotherapy in these patients, with a diagnostic accuracy of 97.4% and 85.2%, respectively.

Introduction

Overexpression of the HER2 gene and its protein is observed in 20% to 30% of breast cancer patients and determines prognosis and treatment. 1 Protein overexpression detected by immunohistochemistry (IHC) or amplification of the HER2 gene analyzed by fluorescence in situ hybridization (FISH) are the methods used to detect HER2 status in clinical practice. 2 However, tumors may contain multiple clones with distinct HER2 amplification characteristics within the tumor, resulting in HER2 intratumoral heterogeneity. Recently, studies have suggested that intratumoral heterogeneity based on HER2 expression levels can affect response to neoadjuvant chemotherapy (NAC). 3-7 The aim of this study was to use MR-based clinical and radiomic features coupled with machine learning to assess tumor heterogeneity in HER2 overexpressing breast cancer patients receiving NAC, and correlate these findings with pathologic response.

Methods

This retrospective single-center study included 311 patients with HER2 overexpressing invasive breast carcinoma who received NAC following pretreatment MRI. Tumor heterogeneity was evaluated based on the HER2 expression: HER2 3+ on IHC; or HER2 less than 3+ on IHC with HER2 gene amplification detected by FISH. Pathologic complete response (pCR) was defined as no residual invasive carcinoma in the breast or axillary lymph nodes (ypT0/isN0). A breast radiologist performed 3D segmentations of the tumor in the first minute post-contrast sequence using ITK-SNAP software. Enhancement maps were calculated as the percentage increase in signal from the pre-contrast image to the first post-contrast image. Radiomics and statistical analysis were performed using MATLAB and publicly available CERR (Computational Environment for Radiological Research) software. For analysis, data was reduced to 16 gray levels and only an interpixel distance of one was considered. CERR analysis results in 102 texture parameters sub-divided into six categories – 22 first order statistics, 26 statistics based on gray level cooccurrence matrices, 16 statistics based on run length matrices, 16 statistics based on size zone matrices, 17 statistics based on neighborhood gray level dependence matrices, and finally 5 statistics based on neighborhood gray tone difference matrices. Initially univariate analysis was performed to identify significant parameters. The 2-tailed Mann Whitney U-test for two independent samples was used to determine significant differences between groups. Correlation analysis was then employed to remove redundant parameters from advancing to model development. If a highly positive (>0.9) or highly negative (<-0.9) correlation was noted the parameter with the lowest AUC from ROC curve analysis was removed. After ROC and correlation analysis, radiomics parameters that demonstrated significant differences between groups were retained and advanced to modelling alongside clinical MR-based parameters, including lesion type (mass/NME/both), multifocality, size and nodal status. Robust machine learning models were developed utilizing coarse decision trees and 5-fold cross-validation.

Results

The mean patient mean age was 49.4 years (range: 24-78 years) and mean tumor size by MRI was 4.7cm (range: 0.9-14.8cm). The index tumor presented as mass in 147 cases (47.3%), non-mass enhancement (NME) in 34 (10.9%) and both mass and NME in 130 (41.8%). Most tumors were multifocal and had suspicious axillary lymph nodes on MRI (65.9% each). 293 tumors (89.6%) were classified as HER2 3+ on IHC and 34 (10.4%) had HER2 gene amplification by FISH. Overall pCR rate was 60.5% (188/311). Statistical analysis resulted in 5 and 8 radiomics parameters advancing for tumor heterogeneity and pathologic response analysis respectively. The four clinical parameters (size, multifocality, type, and nodal status) were also advanced to modelling. The final model to predict HER2 intratumoral expression levels (IHC vs. FISH) utilized 3 MR parameters (2 clinical [lesion type and multifocality] and 1 radiomic [lze]) for a sensitivity of 99.3% (277/279), specificity of 81.3% (26/32), positive predictive value of 97.9% (277/283), negative predictive value of 92.9% (26/28), and diagnostic accuracy of 97.4% (303/311). The final model to predict pCR included 4 MR parameters (2 clinical [lesion type and size] and 2 radiomic [variance and zlv]) for a sensitivity of 85.6% (161/188), specificity of 84.6% (104/123), positive predictive value of 89.4% (161/180), negative predictive value of 79.4% (104/131), and diagnostic accuracy of 85.2% (265/311). These results were independent of age and estrogen-receptor status.

Discussion

Because HER2 status is usually based on the analysis of a small sample of the tumor obtained through needle biopsy before NAC, the development of noninvasive biomarkers to evaluate whole-tumor heterogeneity are needed. Breast MRI is the most accurate imaging method to predict pathologic response in HER2-overexpressing breast cancer after NAC. 8 Radiomics analysis has been shown to enable pathologic response prediction from the pretreatment breast MRI. 9 Imaging intratumoral heterogeneity with MRI delivers additional prognostic value beyond current clinical-pathologic and biologic predictors in breast cancer, and could potentially be used to stratify patients for individualized therapy. 10

Conclusion

The robust machine learning models, including both clinical and radiomics-based MR features, can be used to assess intratumor heterogeneity and predict pCR after NAC in HER2 overexpressing breast cancer patients.

Acknowledgements

This work was partially supported by the NIH/NCI Cancer Center Support Grant (P30 CA008748) and the Breast Cancer Research Foundation

References

  1. Hagemann IS. Molecular testing in breast cancer: A guide to current practices. Arch Pathol Lab Med. 2016;140(8):815–824.
  2. Wolff AC, McShane LM, Hammond MEH, et al. Human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists Clinical Practice Guideline Focused Update. Arch Pathol Lab Med. 2018;142(11):1364–1382.
  3. Hou Y, Nitta H, Wei L, et al. HER2 intratumoral heterogeneity is independently associated with incomplete response to anti-HER2 neoadjuvant chemotherapy in HER2-positive breast carcinoma. Breast Cancer Res Treat. Springer US; 2017;166(2):447–457.
  4. Rye IH, Trinh A, Sætersdal AB, et al. Intratumor heterogeneity defines treatment-resistant HER2+ breast tumors. Mol Oncol. 2018;12(11):1838–1855.
  5. Metzger Filho O, Viale G, Trippa L, et al. HER2 heterogeneity as a predictor of response to neoadjuvant T-DM1 plus pertuzumab: Results from a prospective clinical trial. J Clin Oncol. 2019;37:15_suppl, 502–502.
  6. Muller KE, Marotti JD, Tafe LJ. Pathologic Features and Clinical Implications of Breast Cancer With HER2 Intratumoral Genetic Heterogeneity. Am J Clin Pathol. 2019;7–16.
  7. Krystel-Whittemore M, Xu J, Brogi E, et al. Pathologic complete response rate according to HER2 detection methods in HER2-positive breast cancer treated with neoadjuvant systemic therapy. Breast Cancer Res Treat. 2019; 177:61.
  8. Van Ramshorst MS, Loo CE, Groen EJ, et al. MRI predicts pathologic complete response in HER2-positive breast cancer after neoadjuvant chemotherapy. Breast Cancer Res Treat. Springer US; 2017;164(1):99–106.
  9. Braman NM, Etesami M, Prasanna P, et al. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19(1):1–14.
  10. Wu J, Cao G, Sun X, et al. Intratumoral Spatial Heterogeneity at Perfusion MR Imaging Predicts Recurrence-free Survival in Locally Advanced Breast Cancer Treated with Neoadjuvant Chemotherapy. Radiology. 2018;288(1):26–35.
Proc. Intl. Soc. Mag. Reson. Med. 28 (2020)
4878