Abstract 4188: Evolutionary dynamics in breast cancer via MRI textural kinetic analysis

Objective: Genetic and clinical heterogeneity in breast cancer, as in other malignant tumors, is both a cause and consequence of evolutionary dynamics. Tissue-level intratumoral heterogeneity is apparent on imaging, as magnetic resonance imaging (MRI) studies typically show regional variations in blood flow and cell density. These imaging characteristics can be linked to underlying Darwinian dynamics since blood flow, through delivery of metabolic substrate and growth factors determines key environmental selection forces. In turn, changes in selection forces will select new cellular adaptations and molecular properties. Here we examine the connection between blood flow and a well recognized predictive and prognostic molecular characteristic of breast cancer cells - the expression of estrogen receptor (ER). In a Darwinian environment, cellular properties persist only if they confer a competitive advantage so ER expression will be maintained only if adequate concentrations of estrogen are present. Since estrogen is largely delivered via blood flow, we reasoned that ER negative tumors will be relatively avascular and vice-versa. We hypothesize that, quantitative analysis of blood flow, definable on MRI, can predict the ER status of the cancer cells. Materials and Methods: An IRB approved, retrospective chart review was performed on Breast Imaging Reporting and Data System (BI-RADS) 5 and 6 MRI reports at a single institution from 6/1/2010 to 1/1/2013. Twenty patients with clinical stage IIB or III and histopathologic diagnosis of invasive ductal or lobular carcinoma were identified: 12 were ER positive and 8 ER negative. To quantify blood flow, we examined T1 fat saturated pre and post contrast images performed at five time points: T0 (pre-contrast) and T1-T4 (approximately 120-480 seconds adjusted for breast thickness and field of view). Weight-based Gadolinum contrast agents were utilized. Image analysis included: post initial enhancement (PIE) map from pre and post contrast images, texture features (gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRM) and local binary pattern histogram features (LBP-HF)). The imaging features were compared in ER positive and ER negative tumors. Results: Differing contrast kinetic patterns in ER positive and ER negative breast tumors were observed on textural kinetic analysis. A diagnostic accuracy of 90% was achieved with Naives Bayes classifier and leave-one-out cross validation. The combination of