An Optimal Mass Transport Model to Assess Neoadjuvant Chemotherapy Breast Cancer Response
Abstract
Purpose
To quantify and visualize tumor microenvironment transport behavior in longitudinal breast DCE-MRI acquired during neoadjuvant chemotherapy (NACT), and to develop image-based biomarkers for predicting therapeutic response
Methods
A breast cancer DCE-MRI dataset comprising 153 longitudinal scans from 39 patients was acquired using a standard protocol. Patients underwent ~4 scans: pre-NACT (V1), after the first cycle (V2), mid-treatment (V3), and post-treatment (V4). Surgical pathology classified outcomes as pathologic complete response (pCR; N=8) or non-pCR (N=31). We applied a physics-motivated unbalanced regularized optimal mass transport (urOMT) model, formulated as minimizing a transport cost subject to an advection–diffusion equation with source. For each scan, 12–14 temporal concentration images within a tumor ROI were analyzed. Post-processing produced quantitative metrics (e.g., speed, flux, influx, efflux) and cross-voxel transport trajectories.
Results
(1) urOMT generated 4D maps characterizing advection, diffusion, influx, and efflux, and visualized cross-voxel transport trajectories that highlighted intratumoral heterogeneity. (2) Time-averaged flux, influx, and efflux decreased significantly in pCR compared with non-pCR at V3 and/or V4, consistent with treatment-associated disruption of tumor transport function and supporting these metrics as candidate response biomarkers. The most statistically significant feature was flux at V3 (p = 0.004) and V4 (p = 0.011). (3) Model performance was quantitatively validated, and a key parameter demonstrated robustness on this dataset.
Conclusion
urOMT provides physiologically interpretable, voxel-level measurements of tumor transport properties and trajectory-based visualization of cross-voxel transport. By explicitly incorporating advection and diffusion, it addresses a key limitation of standard voxelwise tracer-kinetic models. Applied to 153 longitudinal breast DCE-MRI during NACT from 39 patients, urOMT-derived transport metrics show potential as imaging biomarkers for therapy monitoring and response assessment.