MRI-Based Mathematical Modeling to Predict the Response of I-SPY 2 Patients with Breast Cancer to Neoadjuvant Therapy

Abstract Purpose: We seek to establish the generalizability of our biology-based mathematical model in accurately predicting the response of patients with locally advanced breast cancer to neoadjuvant therapy (NAT). Experimental Design: Ninety-one patients (representing three subtypes of locally advanced breast cancer) from 10 Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis 2 (I-SPY 2) clinical trial sites received quantitative MRI before (V1), 3 weeks into (V2), and after completion of (V3) the first 12-week standard-of-care or experimental NAT course. We used these data to calibrate, on a patient-specific basis, our previously developed biology-based mathematical model describing the spatiotemporal change in the number of tumor cells. After calibrating the mathematical model to the V1 and V2 MRI data, the calibrated model predicted the patient-specific tumor status at V3 by explicitly accounting for tumor cell movement (constrained by the mechanical properties of the surrounding tissue), proliferation, and death due to treatment. Results: The concordance correlation coefficient between the observed and predicted tumor change from V1 to V3 was 0.94 for total cellularity and 0.91 for volume. A logistic regression model of predicted tumor volume metrics from V1 to V3 differentiated pathologic complete response from nonpathologic complete response patients with an area under the ROC curve of 0.78. Conclusions: Our tumor forecasting pipeline can accurately predict tumor status after an NAT course—on a patient-specific basis, without a training dataset—using “real-world” MRI data obtained from a multi-subtype, multisite clinical trial.