DCE-MRI radiomics features for predicting breast cancer neoadjuvant therapy response

Quantitative MRI plays a central role in the precision management of cancer. Regarding breast cancer (BRCA) patients, DCE-MRI imaging biomarkers (IBs) have shown promising results both in clinical trials and clinical practice. The advancements in the field of radiomics offer more opportunities for defining disease-specific imaging biomarkers for screening, diagnosis and therapy assessment. In this paper we present a study investigating the role of radiomics in predicting breast-cancer therapy response in the neoadjuvant setting. A temporal radiomics approach is proposed for predicting breast cancer therapy response in the neoadjuvant setting on a public cohort of 35 patients with histologically-proven breast cancer of stage II/III with DCE-MRI data available before treatment, after the first cycle and before the end of treatment. The results based on 57 radiomics features indicate that neoadjuvant chemotherapy (NAC) treatment outcome can be predicted both at baseline and right after the first NAC cycle. Our analyses, found that the best predictors were the median and size-zone non-uniformity normalized calculated from the wavelet decomposition of level 2 of the baseline and the first follow-up exam, achieving an AUROC of 80.80% and 81.34% respectively. These encouraging preliminary results call for more relevant research investigating the role of temporal radiomics in predicting NAC outcome towards more personalized therapy planning.