Deep learning prediction of pathological complete response, residual cancer burden, and progression-free survival in breast cancer patients

The goal of this study was to employ novel deep-learning convolutional-neural-network (CNN) to predict pathological complete response (PCR), residual cancer burden (RCB), and progression-free survival (PFS) in breast cancer patients treated with neoadjuvant chemotherapy using longitudinal multiparametric MRI, demographics, and molecular subtypes as inputs. In the I-SPY-1 TRIAL, 155 patients with stage 2 or 3 breast cancer with breast tumors underwent neoadjuvant chemotherapy met the inclusion/exclusion criteria. The inputs were dynamic-contrast-enhanced (DCE) MRI, and T2- weighted MRI as three-dimensional whole-images without the tumor segmentation, as well as molecular subtypes and demographics. The outcomes were PCR, RCB, and PFS. Three (Integrated, Stack and Concatenation) CNN were evaluated using receiver-operating characteristics and mean absolute errors. The Integrated approach outperformed the Stack or Concatenation CNN. Inclusion of both MRI and non-MRI data outperformed either alone. The combined pre- and post-neoadjuvant chemotherapy data outperformed either alone. Using the best model and data combination, PCR prediction yielded an accuracy of 0.81 +/- 0.03 and AUC of 0.83 +/- 0.03; RCB prediction yielded an accuracy of 0.80 +/- 0.02 and Cohen's kappa of 0.73 +/- 0.03; PFS prediction yielded a mean absolute error of 24.6 +/- 0.7 months (survival ranged from 6.6 to 127.5 months). Deep learning using longitudinal multiparametric MRI, demographics, and molecular subtypes accurately predicts PCR, RCB, and PFS in breast cancer patients. This approach may prove useful for treatment selection, planning, execution, and mid-treatment adjustment.

Automated summary

The goal of this study was to use a novel deep-learning convolutional-neural-network (CNN) to predict pathological complete response (PCR), residual cancer burden (RCB), and progression-free survival (PFS) in breast cancer patients treated with neoadjuvant chemotherapy using longitudinal multiparametric MRI, demographics, and molecular subtypes as inputs. The results showed that the Integrated approach of CNN outperformed the Stack or Concatenation CNN, and the combination of MRI and non-MRI data performed better than either alone. The best model achieved an accuracy of 0.81 for PCR prediction, 0.80 for RCB prediction, and a mean absolute error of 24.6 months for PFS prediction.