Abstract 2738: Image-driven biophysical model to predict pathologic response to neoadjuvant therapy in breast cancer

Abstract

Abstract Introduction: In previous work, we developed an image-driven, patient-specific biophysical model of breast tumor response to neoadjuvant therapy (NAT). In this work, we use this model to characterize response during the course of NAT to predict pathological complete response (pCR). pCR is an important binary pathological outcome metric as patients who achieve pCR experience improved survival outcomes. We hypothesize that with optimized imaging time points within the I-SPY 2 trial study design, we will be able to more accurately characterize dynamic response to therapy and predict eventual pCR. Methods: Our retrospective study included 84 patients (26 pCR, 58 non-pCR) from the publicly-available ACRIN-6698 study.4-7 This substudy to the ongoing I-SPY 2 trial evaluates the effectiveness of diffusion weighted magnetic resonance imaging (DW-MRI) for assessing breast cancer response to NAT. We use serial DW-MRI data to compute the apparent diffusion coefficient (ADC), which is used to approximate tumor cell density distribution between baseline and midtreatment (12 weeks). Cellularity data is fit to our model of tumor growth/treatment response1-3 to obtain biophysical estimates of response, including the tumor cell proliferation rate (k). Estimated spatial proliferation maps of the tumor throughout NAT between imaging time points are used to create a histogram of tumor cell proliferation for each patient and histogram metrics are extracted as features to predict pCR. Receiver operating characteristic (ROC) curves were generated and the area under the curve (AUC), sensitivity (sens), and specificity (spec) were calculated to evaluate prediction performance. Results: Parameterizations of tumor biophysical properties were calculated between baseline and midtreatment. In univariate analysis with model-based proliferative area we achieved a ROC AUC of 0.72 (85% sens, 55% spec). Conventional metrics of percent change in tumor area and mean ADC achieved ROC AUC of 0.52 (50% sens and 69% spec; 23% sens and 97% spec). We constructed multivariate logistic regression models to combine parameters to build signatures for predicting pCR. We achieved AUC of 0.88 (85% sens and 81% spec) and 0.97 (100% sens and 88% spec) for our model metrics alone, and in combination with size and ADC metrics, respectively. Conclusions: Use of image-driven mathematical modeling with the ACRIN-6698 subset of the I-SPY 2 trial enables accurate prediction of tumor response during NAT based on baseline and mid-treatment imaging data. This initial study provides promise for improved prediction accuracy with biophysical modeling. These results suggest that a biophysical modeling method has the potential to aid in response characterization. In future work, we will apply our model to the remainder of the ACRIN-6698 subset to more rigorously assess prediction of pCR and investigate incorporation of tumor perfusion and diffusion data. Citation Format: Haley J. Bowers, Jared A. Weis, Alexandra Thomas, Emily Douglas, Katherine Ansley. Image-driven biophysical model to predict pathologic response to neoadjuvant therapy in breast cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 2738.