AUTHOR=Chen Shujun , Shu Zhenyu , Li Yongfeng , Chen Bo , Tang Lirong , Mo Wenju , Shao Guoliang , Shao Feng 

TITLE=Machine Learning-Based Radiomics Nomogram Using Magnetic Resonance Images for Prediction of Neoadjuvant Chemotherapy Efficacy in Breast Cancer Patients

JOURNAL=Frontiers in Oncology

VOLUME=Volume 10 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2020.01410

DOI=10.3389/fonc.2020.01410

ISSN=2234-943X

ABSTRACT=Purpose: The construction and validation of a radiomics nomogram based on machine learning using magnetic resonance image (MRI) for predicting the response of breast cancer patients to neoadjuvant chemotherapy (NACT). 
Methods: This retrospective study consisted of 142 patients with breast cancer who underwent MRI before NACT, of which 35 patients experienced pathological complete response (pCR) by the postoperative pathological examination. The patients were randomly divided into the training (n=99) and test sets (n=43). The features were selected by the maximum relevance minimum redundancy (mRMR) and absolute shrinkage and selection operator (LASSO) algorithm in the training set, which was used to establish the radiomics signature by machine learning. The predictive score of each patient was calculated using the radiomics signature formula. Finally, combined with clinical factors, the predictive model was constructed using multivariate logistic regression, resulting in the creation of the nomogram. The diagnostic accuracy of the nomogram was evaluated by receiver operating characteristics (ROC) analysis and was validated using the test set. Decision curve analysis (DCA) was used to validate the clinical practicability of the nomogram. 
Results: The radiomics signature was well-discriminated in the training set [AUC 0.854, specificity 72%, sensitivity 95.83%], and test set (AUC 0.844, specificity 71.87%, sensitivity 90.91%). The radiomics nomogram, which incorporated the radiomics signature and hormone status, also showed excellent calibration and discrimination in the training set [AUC 0.892, specificity 88.00%, sensitivity 89.17%], and test set (AUC 0.881, specificity 81.25%, sensitivity 86.82%). The decision curve indicated the clinical usefulness of our nomogram.
Conclusion: Our radiomics nomogram showed excellent recognition ability for breast cancer patients with pCR after undergoing neoadjuvant chemotherapy. The model may be used to predict the treatment response of individual breast cancer patients in the future.