Abstract PD2-10: Radiomic phenotypes of tumor heterogeneity from pre-operative DCE-MRI independently predict breast cancer recurrence after 10-year follow-up from primary invasive diagnosis

Abstract

Abstract Purpose: To investigate whether pre-operative MRI phenotypes of tumor heterogeneity have independent prognostic value and determine their associations with recurrence-free-survival (RFS) using 10-year follow-up after primary invasive breast cancer diagnosis. Materials and Methods: The DCE-MRI images of 94 women diagnosed with primary invasive breast cancer, who had complete histopathologic and 10-year follow up data available, were chosen from a previously conducted multimodality clinical trial cohort (2002-2006). For each woman, the signal enhancement ratio (SER) map was calculated for the most representative slice of the primary lesion. Radiomic features (including first order histogram, run-length, structural, and co-occurrence matrix features) and morphologic measures (perimeter, area, ellipticity, and convexity) were extracted and summarized over four quadrants of the tumor. To identify intrinsic phenotypes of tumor heterogeneity, unsupervised hierarchical clustering was applied to the extracted feature vectors after z-score normalization, where cluster cutoffs were determined using Consensus Clustering and the SigClust method. The normalized feature vectors were further averaged to generate a composite heterogeneity score for each tumor. Differences across phenotypes by recurrence, established prognostic factors (ER, PR, HER2, Clinical Stage, Ki67%), and heterogeneity score were assessed using Chi-square and Kruskal- Wallis tests. Kaplan-Meier curves were used to display survival probabilities across phenotypes, adjusting for time-to-event data. Heterogeneity phenotype assignments were added to a baseline Cox proportional hazards model with established prognostic factors to predict RFS. The log-likelihood test was used to assess goodness-of-fit and model discriminative capacity was evaluated using the c-statistic. Results: Our sample included 14 recurrences(15%). Unsupervised clustering identified three intrinsic phenotypes that ranged from low to high heterogeneity (p<0.001). The most heterogeneous phenotype contained all recurrences (p<0.001). Clinical stage was also different among phenotypes (p=0.05). The augmented model incorporating heterogeneity phenotypes had higher discriminatory capacity (c-statistic=0.86), compared to the baseline model including only established histopathologic prognostic factors (c-statistic=0.79). Baseline ModelBaseline Model + MRI Tumor Heterogeneity Phenotype* p= 0.15, c= 0.79p<0.001, c=0.86ParameterHazard RatioHazard RatioEstrogen Receptor (ER)1.001.00Progesterone Receptor (PR)0.990.99HER2 Receptor (Her2)0.650.72Ki67 (%)1.000.99Clinical Stage2.551.94Heterogeneity phenotype p<0.001*Log-likelihood ratio between models p-value <0.001 Conclusion: Intrinsic phenotypes of tumor heterogeneity in pre-operative DCE-MRI of primary invasive breast cancer may provide additional prognostic information to established histopathologic factors. Independent validation is needed to determine phenotype reproducibility and their independent associations to RFS when also including additional molecular/histopathologic prognostic factors. Citation Format: Chitalia RD, Rowland J, McDonald E, Pantalone L, Cohen E, Gastounioti A, Thomas K, Batiste R, Feldman M, Schnall MD, Conant EF, Kontos D. Radiomic phenotypes of tumor heterogeneity from pre-operative DCE-MRI independently predict breast cancer recurrence after 10-year follow-up from primary invasive diagnosis [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; 2017 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2018;78(4 Suppl):Abstract nr PD2-10.