Automated Quantification of Background Parenchymal Enhancement in Longitudinal Dynamic Contrast-Enhanced MRI for Predicting Neo-adjuvant Chemotherapy Response in Breast Cancer

Background: Background parenchymal enhancement (BPE) shows high association with cancer risk and treatment response to neo-adjuvant chemotherapy (NAC) in breast cancer. However, it still lacks automated method for BPE characterization. Methods: In this retrospective study, we ultimately included 894 patients from three cohorts (GDPH, CUM1 and I-SPY2) for analysis. The proposed processing workflow included two main stages: (1) image pre-processing, development of breast and fibroglandular tissue (FGT) segmentation models, (2) BPE index calculation and evaluation. The dice coefficient and area under the curve (AUC) metric were used to evaluate the segmentation performance and discrimination ability of BPE indices for predicting pathological complete response (pCR). Results: For breast segmentation, our model achieved impressive dice scores of 0.88 and 0.90 in the test cohorts. Also, the developed FGT segmentation model attained dice scores 0.86 and 0.86 in the test cohorts respectively, reflecting its proficiency in identifying and isolating fibroglandular tissue regions. The Δ BPE 0-1 indices showed significant association with pCR in the CUM1 and I-SPY2 cohorts, with (OR, 4.861 [CI: 1.248–22.292]; P = 0.030), and (OR, 3.027 [CI: 1.471–6.784]; P = 0.005). Also, the Δ BPE 0-1 index presented better predictive performance, with AUCs of 0.614 (CI: 0.506–0.721) and 0.608 (CI: 0.546–0.671) in the CUM1 and I-SPY2 cohorts respectively. Finally, for early treatment, the BPE indices of sub-cohorts split by molecular subtypes are significantly different in the CUM1 ( P  = 0.044) and I-SPY2 ( P  < 0.001) cohorts respectively. Conclusion: In this study, we proposed a novel workflow to automatically quantify BPE in DCE-MRI, and verified the statistically significant predictive ability of BPE for neo-adjuvant chemotherapy efficacy in multi-center data. The delta-BPE between pre-treatment and early treatment within NAC treatment showed significant association with pCR. It is expected that the developed BPE quantification method can aid personalized treatment for patients with breast cancer.