Abstract PS11-05: Novel 4D radiomics applied to dynamic FES PET images to improve prediction of ER-positive breast cancer outcomes for ER-targeted therapy

Abstract Introduction: Identification of noninvasive biomarkers to predict tumor biology and provide prognostic indicators in ER positive breast cancer are of essence. [18F]fluoroestradiol (FES) is an FDA-approved tracer that measures functional ER expression and predicts response to ER-targeted therapy. The goal of this study was to test novel 4D analysis of dynamic volumetric FES PET patient imaging data to predict outcomes in patients with metastatic ER positive breast cancer undergoing endocrine therapy. Methods: We utilized the Rad-Fit method, previously tested in 4D analysis of dynamic [18F]fluorodexosyglucose (FDG) breast cancer patient data sets, to identify and characterize intratumor subregions through unsupervised clustering of tumor voxels according to each voxel’s time-activity information, summarized by functional principal components (Chitalia, PMID: 3367764). The analysis was performed on a set of dynamic FES PET data taken over a single body section with one or more tumor sites from a previously studied cohort of 38 patients with locally advanced or metastatic breast cancer taken prior to endocrine therapy (Linden, PMID: 1668272). The FES imaging data were analyzed using the Rad-fit method and radiomics measures were compared to patient outcome data (PFS and OS) collected in the original study. Tumors were clustered from two up to twelve clusters and the optimal clustering of each tumor was chosen by utilizing a scaled silhouette score. The mean, standard deviation, minimum, and maximum values of the 4D features of subregion sum of squared error (SSE), distance between sub regions, and the total number of intratumor subregions were used to build prognostic models of overall survival (OS) and progression free survival (PSF). We employed Kaplan-Meyer plots to determine if our prognostic models could significantly separate our population into high and low risk groups. Results: 4D features significantly separated this cohort into a high or low risk group for OS in a single tumor per subject (C=0.73, p=0.0018) or averaging all features per subject (C=0.65, p<0.00025) model. Mean and maximum distance are the most significant radiomic features in the Cox-PH model with all features. Combining the significant radiomic predictors of maximum and mean distance with tumor site SUV Max resulted in improved separation of high and low risk groups for the single tumor (C=0.71, p=0.0014) and average tumor (C=0.61, p=0.00018) models. Conclusions: In this exploratory study, we show that 4D radiomic features extracted from dynamic FES PET images using the established Rad-Fit method can add significantly to prediction of OS in advanced ER positive breast cancer. Metrics of tumor subregion distance, reflective of the overall tumor heterogeneity in FES PET images, appear to perform as the best radiomic predictors for risk stratification of OS. Supported in part by Komen Grant SAC232145 and NIH/NCI Funding R50 CA-211270 Citation Format: Carla Zeballos Torrez, Andrew Chen, Lanell M. Peterson, Mark Muzi, Jennifer M. Specht, Eric A. Cohen, Hannah M. Linden, Despina Kontos, David A. Mankoff. Novel 4D radiomics applied to dynamic FES PET images to improve prediction of ER-positive breast cancer outcomes for ER-targeted therapy [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr PS11-05.