Abstract P5-01-16: Combining HyperVOX with pattern recognition neural networks: A new method for analyzing flow cytometry-based immunophenotyping data for increased early detection of stage I/II breast cancer (BCa)

Abstract

Abstract Breast cancer screening, detection, and diagnosis relies heavily upon mammography, but due to breast pathology and the intrinsic limits of detection, mammogram-based procedures result in many false positives causing an increased risk for overdiagnosis and overtreatment. Therefore, blood-based biomarkers that can accurately predict early stage breast cancer (BCa) in at-risk women are needed. Flow cytometry-based immunophenotyping of peripheral blood is an accessible and non-invasive technology, but as more parameters are included in these panels, new methods must be developed for the efficient analysis and utilization of large sets of flow cytometry data for clinical applications. The goal of this study was to use our newly developed ‘hypervoxelation of cytometry events’ computational technique, or HyperVOX, to transform flow cytometry data into a useable format for input into a series of pattern recognition neural networks (PRNNs) to detect early stage breast cancer (I/II). We used standard multiparametric flow cytometry techniques to immunophenotype myeloid-derived suppressor cell (MDSC), myeloid, and lymphocyte cell populations (14 markers total) found in the peripheral blood of 99 biopsy-confirmed BCa patients (stage I = 70; stage II = 29) with varying hormone receptor expression and 88 healthy donor female (HDF) controls. BCa patients recommended for biopsy by their physician were included in this study and excluded if they had a previous history of cancer or have received any type of treatment for breast cancer in the past 6 months. PRNNs were trained using raw flow cytometry data processed using HyperVOX from 64 BCa patients (stage I = 44; stage II = 20) and 69 HDF controls. Predictions were evaluated using the performance of the trained PRNNs on 35 early stage BCa patients (stage I = 26; stage II = 9) and 19 HDF that were not used for PRNN training (holdout test set). Myeloid-derived suppressor cells (MDSCs) are known to be key contributors in supporting tumor progression and have also been shown to increase in peripheral blood as tumor progression occurs. For this study, manually gated percentages of MDSCs were used for comparison. In the holdout test set, we identified 32 out of 35 subjects with either stage I or II BCa (Sens. = 91.4%, 95%CI = 76.9% to 98.2%; PPV = 97.0%, 95%CI = 82.6% to 99.5%) while 18 out of 19 HDF were identified (Spec. = 94.7%, 95%CI = 74.0% to 99.9%; NPV = 85.7%, 95%CI = 66.9% to 94.7%) for a combined accuracy of 92.6% (95%CI = 82.1% to 97.9%); the resulting AUC = 0.9098 (95%CI = 0.8031 to 1.000). For stage I BCa alone, 24 out of 26 were identified correctly with an AUC = 0.9069 (95%CI = 0.7928 to 1.000). For stage II BCa alone, 8 out of 9 were identified correctly with an AUC = 0.9181 (95%CI = 0.8028 to 1.000). Additionally, we tested 26 samples collected from patients with biopsy confirmed ductal carcinoma in situ (DCIS). Even though they are clinically deemed as pre-cancerous (stage 0), 18 out of 26 (AUC = 0.8421; 95%CI = 0.7163 to 0.9679) were classified as BCa suggesting utility for detecting the existence of even a non-invasive cancerous lesion. Although further study is needed, we believe that this technology, in conjunction with other known clinical risk factors, would allow for clinicians to make a more informed diagnosis and treatment recommendation when screening and for recommending subsequent interventions for early stage breast cancer. Citation Format: George A Dominguez, John Roop, Alexander Polo, Anthony Campisi, Dmitry I Gabrilovich, Amit Kumar. Combining HyperVOX with pattern recognition neural networks: A new method for analyzing flow cytometry-based immunophenotyping data for increased early detection of stage I/II breast cancer (BCa) [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P5-01-16.