Radiomics Analysis of Multi-Phase DCE-MRI in Predicting Tumor Response to Neoadjuvant Therapy in Breast Cancer

Objective: To explore whether the pretreatment dynamic contrast enhancement magnetic resonance imaging (DCE-MRI) and radiomics signatures were associated with pathologic complete response (pCR) to neoadjuvant therapy (NAT) in breast cancer. Method: A retrospective review of 70 patients with breast invasive carcinomas proved by biopsy between June 2017 and October 2020 (26 patients were pathological complete response, and 44 patients were non-pathological complete response). Within the pre-contrast and five post-contrast dynamic series, a total of 1037 quantitative imaging features were extracted from in each phase. Additionally, the & UDelta;features (the difference between the features before and after the comparison) were used for subsequent analysis. The least absolute shrinkage and selection operator (LASSO) regression method was used to select features related to pCR, and then use these features to train multiple machine learning classifiers to predict the probability of pCR for a given patient. The area under the curve (AUC), accuracy, sensitivity, and specificity were calculated to assess the predictive performances of the radiomics model for each of the five phases of time points. Result: Among the five phases, each individual phase performed with AUCs ranging from 0.845 to 0.919 in predicting pCR. The best single phases performance was given by the 3rd phase (AUC = 0.919, sensitivity 0.885, specificity 0.864). 5 of the features have significant differences between pCR and non-pCR groups in each phase, most features reach their maximum or minimum in the 2nd or 3rd phase. Conclusion: The radiomic features extracted from each phase of pre-treatment DCE-MRI possess discriminatory power to predict tumor response.

Automated summary

This study found that radiomic features extracted from each phase of pre-treatment DCE-MRI can predict tumor response in breast cancer patients undergoing neoadjuvant therapy. Using machine learning classifiers and feature selection methods, the probability of pCR in patients can be predicted effectively.