Treatment Response Prediction Using Ultrasound-Based Pre-, Post-Early, and Delta Radiomics in Neoadjuvant Chemotherapy in Breast Cancer

Min Yang; Huan Liu; Qingli Dai; Ling Yao; Shun Zhang; Zhihong Wang; Jing Li; Qinghong Duan

doi:10.3389/fonc.2022.748008

Treatment Response Prediction Using Ultrasound-Based Pre-, Post-Early, and Delta Radiomics in Neoadjuvant Chemotherapy in Breast Cancer

Front Oncol. 2022 Feb 7:12:748008. doi: 10.3389/fonc.2022.748008. eCollection 2022.

Authors

Min Yang¹, Huan Liu², Qingli Dai¹, Ling Yao¹, Shun Zhang¹, Zhihong Wang³, Jing Li⁴, Qinghong Duan¹

Affiliations

¹ Department of Medical Imaging, the Affiliated Tumor Hospital of Guizhou Medical University, Guiyang, China.
² Department of Advanced Application Team, GE Healthcare, Shanghai, China.
³ Department of Breast Surgery, the Affiliated Tumor Hospital of Guizhou Medical University, Guiyang, China.
⁴ Department of Medical Imaging, The Affiliated Hospital of Guizhou Medical University, Guiyang, China.

Abstract

Objective: To develop and validate a radiomics nomogram based on pre-treatment, early treatment ultrasound (US) radiomics features combined with clinical characteristics for early prediction of response to neoadjuvant chemotherapy (NAC) in breast cancer.

Method: A total of 217 patients with histological results of breast cancer receiving four to eight cycles of NAC before surgery from January 2018 to December 2020 were enrolled. Patients from the study population were randomly separated into a training set (n = 152) and a validation set (n = 65) at a ratio of 7:3. A total of 788 radiomics features were extracted from each region of interest in the US image at pre-treatment baseline (radiomic signature, RS1), early treatment (after completion of two cycles of NAC, RS2) and delta radiomics (calculated between the pre-treatment and post-treatment features, Delta RS). The Max-Relevance and Min-Redundancy (mRMR) and the least absolute shrinkage and selection operator (LASSO) regression were applied for feature selection. The predictive nomogram was built based on the radiomics signature combined with clinicopathological risk factors. Discrimination, calibration, and prediction performance were further evaluated in the validation set.

Results: Of the 217 breast masses, 127 (58.5%) were responsive to NAC and 90 (41.5%) were non-responsive. Following feature selection, nine features in RS1, 11 features in RS2, and eight features in Delta RS remained. With multivariate analysis, the RS1, RS2, Delta RS, and Ki-67 expression were independently associated with breast NAC response. However, the performance of the Delta RS (AUC _{Delta RS} = 0.743) was not higher than RS1 (AUC _RS1 = 0.722, P_{Delta vs RS1} = 0.086) and RS2 (AUC _RS2 = 0.811, P_{Delta vs RS2 =} 0.173) with the Delong test. The nomogram incorporating RS1, RS2, and Ki-67 expression showed better predictive ability for NAC response with an area under the curve (AUC) of 0.866 in validation cohorts than either the single RS1 (AUC 0.725) or RS2 (AUC 0.793) or Ki-67 (AUC 0.643).

Conclusion: The nomogram incorporating pre-treatment and early-treatment US radiomics features and Ki-67 expression showed good performance in terms of NAC response in breast cancer, thereby providing valuable information for individual treatment and timely adjustment of chemotherapy regimens.

Keywords: Ki-67; breast cancer; neoadjuvant chemotherapy; nomogram radiomics; ultrasound.