Imaging plays a critical role in evaluating tumor response to treatment; however the currently used methods remain significantly limited. For example, standards such as the RECIST are subjective and cannot be used to adequately characterize irregular lesions; tumor volume measures alone do not account for detailed structural changes; and features from selected tumor regions, such as hot-spot peak-enhancement, do not capture information from the entire tumor. As such, current approaches fall short of capturing the multi-faceted effects of treatment, including phenotypic tumor heterogeneity and its longitudinal change during treatment, which is increasingly recognized as an important predictive indicator. To date, few studies have explored using richer imaging descriptors, which could result in more powerful predictive markers. Moreover, fewer have attempted to combine multi-modal biomarkers, such as imaging with histopathologic and molecular markers, to develop enhanced predictive models for specific tumor sub-types and individual patients. We propose to develop advanced computational tools that will enable to i) extract novel multi-parametric imaging signatures and ii) accurately characterize their longitudinal patterns of change during neoadjuvant treatment via deformable image registration. Our approach is thus geared towards knowledge discovery, for determining which imaging parameters have the highest predictive value out of many possible ways to quantify information provided by imaging. In SA1 we will develop robust 4D deformable image registration methods, based on principles of mutual saliency, for estimating transformations that will enable us to robustly register serial imaging scans and obtain anatomically precise spatio-temporal parametric maps of longitudinal tissue effects induced by treatment. In SA2 we will analyze whole-tumor and normal tissue effects by performing multi- parametric feature extraction, including a rich set of morphologic, textural, kinetic and parenchymal tissue descriptors, which in conjunction to registration will allow us to comprehensively capture the dynamically evolving imaging phenotype during treatment. In SA3 we will test our method in a major breast imaging study, the I-SPY 1/ACRIN 6657 trial. We will apply machine learning tools to identify high-dimensional associations of imaging patterns, in conjunction to histopathologic tumor subtyping, that can best predict pathologic complete response (pCR) and 5-year disease free survival (DFS). In SA4 we will independently test our models with the I-SPY 2/ACRIN 6698 trial, where we will also evaluate the robustness of our features to a diverse range of treatments. Our methods hold the promise to shift the current paradigm in personalizing neoadjuvant treatment by 1) improving the current standards of imaging-based assessment and 2) introducing new imaging biomarkers that can be of higher value as early predictors of treatment response and survival. Our tools will be shared as open-source software via NIH/NCI tool registries and open-challenge activities.

Public Health Relevance

Pattern analysis and machine learning methods are powerful knowledge discovery tools that can be very effective in identifying complex multi-parametric patterns from a diverse set of imaging and non-imaging biomarkers that can best predict treatment response and outcome. Our method will leverage the richness of available data to point to new imaging biomarkers that are better early predictors of response, enabling personalized treatment decisions for women undergoing neoadjuvant chemotherapy for breast cancer.

National Institute of Health (NIH)
National Cancer Institute (NCI)
Research Project (R01)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1)
Program Officer
Baker, Houston
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Pennsylvania
Schools of Medicine
United States
Zip Code
Davatzikos, Christos; Rathore, Saima; Bakas, Spyridon et al. (2018) Cancer imaging phenomics toolkit: quantitative imaging analytics for precision diagnostics and predictive modeling of clinical outcome. J Med Imaging (Bellingham) 5:011018
Ashraf, Ahmed; Gaonkar, Bilwaj; Mies, Carolyn et al. (2015) Breast DCE-MRI Kinetic Heterogeneity Tumor Markers: Preliminary Associations With Neoadjuvant Chemotherapy Response. Transl Oncol 8:154-62
Ou, Yangming; Weinstein, Susan P; Conant, Emily F et al. (2015) Deformable registration for quantifying longitudinal tumor changes during neoadjuvant chemotherapy. Magn Reson Med 73:2343-56