Detecting Prostate Cancer Using Pattern Recognition Neural Networks With Flow Cytometry-Based Immunophenotyping in At-Risk Men

Abstract

Current screening methods for prostate cancer (PCa) result in a large number of false positives making it difficult for clinicians to assess disease status, thus warranting advancements in screening and early detection methods. The goal of this study was to design a liquid biopsy test that uses flow cytometry-based immunophenotyping and artificial neural network (ANN) analysis to detect PCa. Numerous myeloid and lymphoid cell populations, including myeloid-derived suppressor cells, were measured from 156 patients with PCa, 123 with benign prostatic hyperplasia (BPH), and 99 male healthy donor (HD) controls. Using pattern recognition neural network (PRNN) analysis, a type of ANN, PCa detection compared against HD resulted in 96.6% sensitivity, 87.5% specificity, and an area under the curve (AUC) value of 0.97. Detecting patients with higher risk disease (⩾Gleason 7) against lower risk disease (BPH/Gleason 6) resulted in 92.0% sensitivity, 42.7% specificity, and an AUC of 0.72. This study suggests that analyzing flow cytometry immunophenotyping data with PRNNs may prove to be a useful tool to improve PCa detection and reduce the number of unnecessary prostate biopsies performed each year.

Keywords: Liquid biopsy; early detection; machine learning; neural network; prostate cancer.

Figure 1.

Comparisons of levels of different subsets of circulating myeloid and lymphoid cells between HDM, patients with BPH, and patients with PCa. Peripheral blood from 99 HDM, 123 patients with BPH, and 156 patients with PCa were stained for myeloid and lymphoid cell markers. Scatter plots show the median of calculated percentages of live cells from each population with interquartile ranges of (A) eMDSCs, (B) PMN-MDSCs, (C) M-MDSCs, (D) CD16^‒PMN, (E) CD14⁺ monocytes, (F) classical monocytes, (G) nonclassical monocytes, (H) CD3⁺ lymphocytes, (I) T cells, (J) CD4⁺ T cells, (K) CD8⁺ T cells, (L) natural killer T (NKT) cells, (M) CD4⁺ NKT cells, (N) CD8⁺ NKT cells, (O) NK cells, and (P) B cells. (Q) ROC curves determined from calculated percentages of live cells from the 3 MDSC populations for all patients with PCa versus HDM. HDM n = 99, BPH n = 123, and PCa n = 156. BPH, benign prostatic hyperplasia; eMDSCs, early-stage MDSCs; HDM, male healthy donors; MDSCs, myeloid-derived suppressor cells; M-MDSCs, monocytic MDSCs; PCa, prostate cancer; PMN, polymorphonuclear. *P < .05; **P < .005; ***P < .0005; ****P < .0001.

Figure 2.

Comparisons of levels of different subsets of circulating myeloid and lymphoid cells between HDM, patients with BPH, and patients with PCa. Peripheral blood from 99 HDM, 123 BPH, 59 Gleason 6 (G6) PCa, 68 Gleason 7 (G7) PCa, and 29 >Gleason 7 (>G7) PCa patients were stained for myeloid and lymphoid markers. Scatter plots show the median of calculated percentages with interquartile ranges of (A) eMDSCs, (B) PMN-MDSCs, (C) M-MDSCs, (D) CD16^‒PMN, (E) CD14⁺ monocytes, (F) classical monocytes, (G) nonclassical monocytes, (H) CD3⁺ lymphocytes, (I) T cells, (J) CD4⁺T cells, (K) CD8⁺T cells, (L) natural killer T (NKT) cells, (M) CD4⁺NKT cells, (N) CD8⁺NKT cells, (O) NK cells, and (P) B cells. BPH, benign prostatic hyperplasia; eMDSCs, early-stage MDSCs; HDM, male healthy donors; MDSCs, myeloid-derived suppressor cells; M-MDSCs, monocytic MDSCs; PCa, prostate cancer; PMN, polymorphonuclear. *P < .05; **P < .005; ***P < .0005; ****P < .0001.

Figure 3.

Heat map of the normalized median percentages for all myeloid and lymphoid cell populations. The numbers in the heat map represent the normalized percentage values to either HDM or patients with BPH. Increases in value are shaded in red and decreases in value are shaded in blue. (A) Median percentage values of all cell populations from BPH, G6 PCa, G7(3 + 4) PCa, G7(4 + 3) PCa, and >G7 PCa patients normalized to their respective HDM cell population value. (B) Median percentage values of all cell populations from G6 PCa, G7(3 + 4) PCa, G7(4 + 3) PCa, and >G7 PCa patients normalized to their respective BPH cell population value. BPH, benign prostatic hyperplaisa; PCa, prostate cancer; HDM, male healthy donors; G6, Gleason score 6; G7(3 + 4), Gleason score 7 (primary Gleason score 3 + secondary Gleason score 4) PCa; G7 (4 + 3), Gleason score 7(primary Gleason score 4 + secondary Gleason score 3) PCa; >G7, Gleason score >7.

Figure 4.

Neural network analysis of hypervoxelated flow cytometry data of hold out samples for patients with PCa versus HDM and PCa versus BPH. (A) Percentages of correctly classified samples from HDM (n = 32), G6 PCa (n = 29), G7(3 + 4) PCa (n = 10), G7(4 + 3) PCa (n = 9), G8 PCa (n = 4) and G9 PCa (n = 7) and (B) ROC curve for NN analysis for holdout samples of patients with PCa versus HDM. (C) Percentages of correctly classified samples from BPH (n = 87), G6 PCa (n = 23), G7(3 + 4) PCa (n = 2), G7(4 + 3) PCa (n = 17), G8 PCa (n = 2) and G9 PCa (n = 4) and (D) ROC curve for NN analysis for holdout samples of patients with PCa versus BPH. G6, Gleason score 6 PCa; G7 (3 + 4), Gleason score 7 (primary Gleason score 3 + secondary Gleason score 4) PCa; G7(4 + 3), Gleason score 7 (primary Gleason score 4 + secondary Gleason score 3) PCa; G8, Gleason score 8 PCa; G9, Gleason score 9 PCa. BPH, benign prostatic hyperplasia; HDM, male healthy donors; NN, neural network; PCa, prostate cancer; ROC, receiver operating characteristic.