Detection of Ovarian Cancer through Exhaled Breath by Electronic Nose: A Prospective Study

Abstract

Background: Diagnostic methods for the early identification of ovarian cancer (OC) represent an unmet clinical need, as no reliable diagnostic tools are available. Here, we tested the feasibility of electronic nose (e-nose), composed of ten metal oxide semiconductor (MOS) sensors, as a diagnostic tool for OC detection.

Methods: Women with suspected ovarian masses and healthy subjects had volatile organic compounds analysis of the exhaled breath using e-nose.

Results: E-nose analysis was performed on breath samples collected from 251 women divided into three groups: 86 OC cases, 51 benign masses, and 114 controls. Data collected were analyzed by Principal Component Analysis (PCA) and K-Nearest Neighbors' algorithm (K-NN). A first 1-K-NN (cases vs. controls) model has been developed to discriminate between OC cases and controls; the model performance tested in the prediction gave 98% of sensitivity and 95% of specificity, when the strict class prediction was applied; a second 1-K-NN (cases vs. controls + benign) model was built by grouping the non-cancer groups (controls + benign), thus considering two classes, cases and controls + benign; the model performance in the prediction was of 89% for sensitivity and 86% for specificity when the strict class prediction was applied.

Conclusions: Our preliminary results suggested the potential role of e-nose for the detection of OC. Further studies aiming to test the potential adoption of e-nose in the early diagnosis of OC are needed.

Keywords: K-NN models; MOS sensors; early diagnosis; electronic nose; ovarian cancer.

Figure 1

Metal oxide semiconductor (MOS) sensor signals for a case (a) and control (b) sample. The histograms represent the responses after 40 s of sampling for the 10 e-nose sensors (S1–10).

Figure 2

Interval plot of electronic nose data set for the ovarian cancer (OC) cases (red) and controls (green).

Figure 3

Principal Component Analysis(PCA) score plots (a1,a2) and loading plot (b) of electronic nose data collected on control (green), benign (yellow), and case (red) samples. In the score plot (a1) empty symbols are for premenopausal (pre-m) patients, and full symbols are for postmenopausal (post-m) patients. In the score plot (a2) empty black symbols are for early-stage patients. Two score pots were elaborated to: i. discriminate premenopausal and postmenopausal patients (Figure 3(a1)). ii. evaluate the plot distribution for early-stage ovarian cancer (OC) patients (Figure 3(a2)).