Multiparametric MRI-based longitudinal-radiomics analysis for early prediction of treatment response of breast cancers to neoadjuvant chemotherapy.

[PURPOSE] Our study examined the capability of longitudinal radiomics, based on multi-parametric MRI and validated with a radiomics strategy, to predict how breast cancer responds to neoadjuvant chemotherapy.

[METHODS AND MATERIALS] Longitudinal radiomics analysis was conducted using 3D Slicer software on dynamic contrast-enhanced, T2-weighted, and diffusion-weighted magnetic resonance imaging (MRI) before and during neoadjuvant therapy (after three to four courses). PyRadiomics (V3.0.1) was used to extract radiomics features from both unaltered and filtered images. The filters applied were Gaussian Laplacian and wavelet. A stacking strategy was used, which included random forest (RF), multilayer perceptron (MLP), logistic regression (LR), linear discriminant analysis (LDA), extreme gradient boosting (XGB) and support vector machine (SVM, radial basis function), to integrate the results from the base models and offer secondary prediction outputs.

[RESULTS] In distinguishing between Residual Cancer Burden (RCB) 0 and RCB 1-3, the longitudinal radiomics model achieved an area under the receiver operating characteristic curve (AUC) of 0.909 in the training set and 0.893 in the test set. When identifying RCB 3 and RCB 0-2, the maximum AUC of the longitudinal radiomics model in the training and test sets was 0.981 and 0.916, respectively.

[CONCLUSION] In summary, our research results exhibit good predictive power for RCB assessment. In the future, they will assist clinicians in guiding individualized treatment plans for breast cancer patients undergoing NAC.