Radiomics analysis of contrast-free synthetic MRI and apparent diffusion coefficient for early predicting neoadjuvant...

To investigate the value of the radiomics features extracted from the contrast-free synthetic MRI (SyMRI) and apparent diffusion coefficient (ADC) in the early prediction of treatment response to neoadjuvant chemotherapy (NAC) in breast cancer.

267 breast cancer patients who received neoadjuvant chemotherapy (NAC) before surgery were prospectively enrolled and divided into the training cohort (n = 145), validation cohort (n = 62) and test cohort (n = 60) in a chronological order. All patients underwent breast MRI with SyMRI and Diffusion-weighted imaging (DWI) before (pre-) and after the first cycle (1st-) of NAC. Pre- and 1st- radiomics features were extracted from DWI-ADC and SyMRI quantitative maps (SyT1/SyT2/SyPD), and their changes (delta-radiomics features) were calculated. The support vector machine with recursive feature elimination (SVM-RFE) was used to select features and build models using separate and combined maps. The model performance was assessed by area under receiver operating characteristic curve (AUC) and compared by Delong test.

SyMRI-derived delta-radiomics models outperformed the corresponding pre- and 1st-models. The 1st-ADC radiomics model outperformed the corresponding pre- and delta-models. The delta-SyT2 and 1st-ADC achieved the optimal combination with a higher but no significant AUC (training/validation/test) of 0.809(95% CI: 0.730-0.888)/0.809(95% CI: 0.691-0.927)/0.810 (95% CI: 0.687-0.933) compared with the separate models. After adding receptor status, the noncontrast clinical-radiomics achieved the highest AUC of 0.916(95% CI: 0.869-0.963)/0.914(95% CI: 0.838-0.991)/0.850(95% CI: 0.747-0.954) compared with the clinical and radiomics models.

Multiparametric SyMRI and ADC, as noncontrast sequences, have the potential to early predict NAC response in radiomics analysis, thus reducing the need for repeated contrast agent injections during treatment.