Abstract 3144: Using artificial intelligence and flow cytometry to confirm the need for a prostate biopsy in at-risk men

Abstract

Abstract The goal of this study was to create a blood based, liquid biopsy test for confirming the need for a prostate biopsy by objectively analyzing flow cytometry-based immunophenotyping data using machine learning (ML) by evaluating whether a patient is at-risk for having intermediate-/high-grade prostate cancer PCa [Gleason score (GS) ≥ 7]. Currently, there is a lack of blood-based biomarkers that can accurately predict high-risk PCa in at-risk men making it difficult for clinicians to assess disease status. Furthermore, a large percentage of men undergo unnecessary prostate biopsy procedures each year since the majority of biopsies return either benign pathologies, such as benign prostatic hyperplasia (BPH), or low-grade PCa [Gleason score (GS) = 6). Flow cytometry-based immunophenotyping of peripheral blood is an accessible and non-invasive technology, but as more biomarkers are included, new methods must be developed for the efficient analysis and utilization of these large datasets for clinical applications. Here, we used ML to create pattern recognition neural networks (PRNNs) to identify patients with intermediate-/high-grade PCa from those that have either low-grade PCa or BPH based upon the immunophenotyping of 27 different myeloid and lymphoid cell populations. A key population of interest are myeloid-derived suppressor cells (MDSCs) since they are known to be key contributors in supporting tumor progression and escape through their ability to suppress anti-tumor responses mediated through T cell and natural killer (NK) cell activity. We used standard multiparametric flow cytometry to measure MDSCs and other immune cell populations found in the peripheral blood of 114 biopsy-confirmed PCa and 58 biopsy-confirmed BPH patients that were collected in Streck Cyto-Chex® BCT. PRNNs were trained with raw flow cytometry from two data sets: 37 patients with PCa GS ≥ 7 and 40 patients with BPH/PCa GS6. Predictions were evaluated using the performance of the trained PRNNs on known PCa GS ≥ 7 and BPH/PCa GS6 not used in the PRNN training set (holdout samples). In the holdout samples, we identified 21 out of 23 patients who were identified as PCa GS ≥ 7 (Sens = 91.3%, 95%CI = 72.0% to 98.9%; PPV = 33.3%, 95%CI = 28.4% to 38.7%) while 30 out of 72 patients were identified as BPH/PCa GS6 (Spec = 41.7%, 95%CI = 30.2% to 53.9%; NPV = 93.8%, 95%CI = 79.5% to 98.3%) for a combined accuracy of 53.7% (95%CI = 43.2% to 64.0%). The PRNN classified 17 out of 33 BPH, 13 out of 26 GS6 PCa, 13 out of 14 GS7 PCa, and 8 out of 9 > GS7 PCa samples correctly. For this study, out of the 112 patients who were ultimately diagnosed with BPH or low-grade PCa, approximately 47 procedures could have been avoided or delayed. In a clinical setting, we believe that this technology, in use with other known clinical factors, would allow for clinicians to have a more informed decision when recommending their patients for a prostate biopsy procedure while reducing the number of unnecessary biopsy procedures performed each year. Citation Format: George A. Dominguez, John Roop, Anthony J. Campisi, Thomas Schlumpberger, Amit Kumar. Using artificial intelligence and flow cytometry to confirm the need for a prostate biopsy in at-risk men [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3144.