Spatiotemporal modeling with SimBioSys TumorScope to predict chemotherapeutic response in breast tumor microenvironments.

Background: Predictive models of the efficacy of different tumor therapies will provide significant enhancements to current standard of care practices. Predicting a given tumor’s growth and treatment response, however, is an intricate process that requires not only an understanding of the tumor’s intrinsic characteristics, but also spatial- and temporal-resolved tumor shape descriptions, surrounding tissue dynamics, and a complete account for the milieu of diffusible molecules that drive tumor behaviors and interactions. Here we report an ongoing retrospective study designed to validate SimBioSys TumorScope as a computational tumor therapy prediction model in a real-world clinical setting.

Methods: Fully-deidentified and HIPAA-compliant data were assessed from real-world clinical records and cases. Subjects comprised early stage breast cancer patients who were treated with neoadjuvant chemotherapy (NACT) and subsequent surgical resection. Data fields included imaging data, biomarker status, tumor sizing, demographic data, digital pathology, and genetic lab test data. Half of the data, including all data fields, were used as a training dataset for TumorScope. The second half of the data, with blinded diagnoses and results, will be used to test TumorScope’s prediction accuracy. Simulations will be initialized with pre-treatment MRI data and processed through the entirety of each patient’s specified treatment regimen. Predicted tumor volume and longest dimension will be compared against measured values at several time-points after therapy initiation. Overall accuracy will be statistically assessed by the Pearson correlation coefficient between predicted and measured tumor volume and longest dimension at each time-point, as well as their root-mean-squared-errors.

Results: Final statistical analysis is currently underway. Thus far, SimBioSys TumorScope has trended high accuracy levels with the non-blinded “training” cohort just as it has in previous database studies, with a Pearson correlation coefficient greater than 0.94.

Conclusions: SimBioSys TumorScope for Breast Cancer accurately predicts patient NACT responses via spatio-temporal modeling of drug and nutrient perfusion, metabolic behavior, and the physio-chemical interactions between surrounding tissues. Future prospective studies may assess TumorScope’s capacity to support efficient patient treatments and enhance overall standard of care.

View full publication and poster here.