Over 20,000 patients in the US annually undergo neo-adjuvant chemotherapy (NAC) prior to surgery as a standard-of-care treatment for locally advanced breast cancers, but 70-90% will ultimately fail to achieve a complete response. If identified prior to treatment, patients who will respond poorly to standard NAC could be immediately placed on more aggressive treatment regimens, circumventing an ineffective treatment window that introduces unnecessary suffering and increases risk of progression. While tumor changes throughout treatment can be monitored using clinical dynamic contrast-enhanced (DCE) MRI, there remains no clinically-accepted pre-treatment predictors of NAC response. Radiomic analysis, defined as high-throughput extraction of quantitative image features, has been demonstrated to enable earlier prediction of response from DCE-MRI with improved accuracy. However, most of the features are limited to texture and shape features of the nodule, ignoring the opportunity to interrogate features of the tumor microenvironment known to be implicated in treatment response. Furthermore, a critical roadblock in the wide-scale adoption of radiomics for treatment response prediction is its low biological interpretability, as its features lack an established molecular and morphologic basis. Radiogenomic approaches, which seek to identify connections between imaging and molecular markers of disease, provide greater biologic intuition, but often without application to clinical outcomes. We propose a systems-biology based approach to predict NAC response from baseline breast DCE-MRI with high clinical interpretability and robustness. We will develop novel radiomic features targeted to response- associated tumor biology in the tumor and tumor microenvironment (e.g. immune response and angiogenesis), then validate both their capability to predict therapeutic outcomes and their basis in multi-scale tumor biology. Ultimately, the research proposed could provide effective, non-invasive guidance of NAC without sacrificing biological interpretability of the features, an important pre-requisite for clinical adoption of these tools.
Aim 1 will seek to develop a set of systems-biology driven radiomic descriptors to characterize the biology within the tumor and its microenvironment from breast DCE-MRI. We have previously shown that texture heterogeneity features in the peri-tumoral microenvironment on post-contrast MRI are predictive of treatment response and associated with immune response. We will expand these features to capture temporal changes in 3D heterogeneity, both intra- and peri-tumorally. Additionally, we will develop morphological features to characterize organization of the tumor-associated vascular network.
Aim 2 will focus on discovering and validating imaging signatures from baseline DCE-MRI which are predictive of response to NAC with and without HER2-targeted therapy.
Aim 3 will elucidate the molecular and morphological basis of predictive radiomic features identified in Aim 2 by exploring their associations with aberrations on the genomic and histology scales (immune response and angiogenesis) from pre-treatment biopsy samples.
Developing radiogenomic tools for the pre-treatment prediction of response to neo- adjuvant chemotherapy enable the non-invasive guidance of therapeutic approach without requiring an initial treatment period. The Imaging-based Neoadjuvant Response Score will assist clinicians in stratifying breast cancer patients who will (i) have pathological complete response to treatment and benefit most and (ii) will not benefit from standard therapy and can be immediately moved to more aggressive treatment.
