Diagnosis of head-and-neck cancer from exhaled breath

Abstract

Background: Head-and-neck cancer (HNC) is the eighth most common malignancy worldwide. It is often diagnosed late due to a lack of screening methods and overall cure is achieved in <50% of patients. Head-and-neck cancer sufferers often develop a second primary tumour that can affect the entire aero-digestive tract, mostly HNC or lung cancer (LC), making lifelong follow-up necessary.

Methods: Alveolar breath was collected from 87 volunteers (HNC and LC patients and healthy controls) in a cross-sectional clinical trial. The discriminative power of a tailor-made Nanoscale Artificial Nose (NA-NOSE) based on an array of five gold nanoparticle sensors was tested, using 62 breath samples. The NA-NOSE signals were analysed to detect statistically significant differences between the sub-populations using (i) principal component analysis with ANOVA and Student's t-test and (ii) support vector machines and cross-validation. The identification of NA-NOSE patterns was supported by comparative analysis of the chemical composition of the breath through gas chromatography in conjunction with mass spectrometry (GC-MS), using 40 breath samples.

Results: The NA-NOSE could clearly distinguish between (i) HNC patients and healthy controls, (ii) LC patients and healthy controls, and (iii) HNC and LC patients. The GC-MS analysis showed statistically significant differences in the chemical composition of the breath of the three groups.

Conclusion: The presented results could lead to the development of a cost-effective, fast, and reliable method for the differential diagnosis of HNC that is based on breath testing with an NA-NOSE, with a future potential as screening tool.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1

PCA plots of the PC1 and PC2 values of the five sensor NA-NOSE responses of (A) HNC and healthy sub-populations, (B) LC and healthy sub-populations, (C) HNC and LC, and (D) all patients: HNC, LC, and healthy controls. Each patient is represented by one point in plot. The first two principal components depicted contained 80, 67 and 70 and 66% for (A–D), respectively, of the total variance in the data. All test persons including the misclassified were considered in the statistical analysis.

Figure 2

PCA of the PC1 and PC2 values resulting from statistical analysis of the abundance of volatile biomarkers determined by GC–MS/SPME analysis, using (A) six common volatile biomarkers for distinguishing HNC from healthy states; (B) seven common volatile biomarkers to distinguish HNC from LC. The compound names, masses, and CAS registry numbers are listed in the tables on the right of the PC plots.