2. Differential Diagnosis and Disease Phenotyping and Staging in Breath Analysis
As can be observed on
Table 2, exhaled VOCs have been used not only for disease diagnosis in comparison to healthy subjects, but also for the discrimination between different diseases (e.g., GCa, LC, and asthma/COPD
[4344]) or even for the discrimination between the different stages of a particular disease (e.g., CKD stages
[4647]). One of the main prerequisites for the development of clinically applicable diagnostic tests is the effective discrimination between different diseases with similar symptoms and biochemical pathways
[4748]. The uncertainty in the differentiation of patients with distinct diseases comprises one of the main drawbacks for studies distinguishing a specific disease from HC
[4849]. Thus, the use of breath analysis for differential diagnosis as well as disease staging or phenotyping, using either analytical techniques or sensing devices (
Table 3), attracted significant research interest over the last years. As expected, the contribution of nanomaterials is of great importance with many recent publications of nanomaterial-based sensors focusing on disease differentiation, staging or phenotyping, rather than the simple discrimination between patients and HC (
Table 3).
Table 2. Sensor arrays for exhaled VOCs detection as biomarkers of several diseases, in real-world or synthetic samples, using conventional materials and/or nanomaterials.
Sensing Element |
Disease |
Targeted VOCs |
LOD |
Subjects |
Classifier |
Results |
T |
Ref. |
In Vivo Studies—Real-World Samples |
Chemiresistor—arrays |
Molecularly capped AuNPs—14 different ligands |
Lung cancer |
1-Methyl-4-(1-methyl ethyl) benzene, Toluene, 3,3-Dimethyl pentane, 2,3,4-Trimethyl hexane, Dodecane, 1,1′-1-Butenylidene)bis benzene |
NA |
30 LC, 26 CC, 22 BC, 18 PC, 22 HC |
PCA |
Good discrimination of cancer types from HC, but not between them. No VOC overlap in abundance between HC and cancer patients. |
RT |
[16] |
Colorectal cancer |
1,1′-(1-Butenylidene)bis benzene, 1,3-Dimethyl benzene, 1-Iodo nonane, (1,1-Dimethylethyl)thio acetic acid, 4-(4-Propylcyclohexyl)-4′-cyano1,1′-biphenyl-4-yl ester benzoic acid, 2-Amino-5-isopropyl-8-methyl-1-azulene carbonitrile |
Breast cancer |
3,3-Dimethyl pentane, 2-Amino-5-isopropyl-8-methyl-1-azulene carbonitrile, 5-(2-Methylpropyl)nonane, 2,3,4-Trimethyl decane, 6-Ethyl-3-octyl ester 2-trifluoromethyl benzoic acid |
Prostate cancer |
Toluene, 2-Amino-5-isopropyl-8-methyl-1-azulene carbonitrile, 2,2-Dimethyl decane, p-Xylene |
Molecularly capped AuNPs—7 different ligands |
Prostate cancer |
Toluene, 2-Amino-5-isopropyl-8-methyl-1-azulene carbonitrile, 2,2-Dimethyl decane, p-Xylene |
NA |
9 PC, 10 HC |
DFA |
100% specificity, 100% sensitivity |
RT |
[4243] |
Breast cancer |
3,3-Dimethyl pentane, 2-Amino-5-isopropyl-8-methyl-1-azulene carbonitrile, 5-(2-Methylpropyl)nonane, 2,3,4-Trimethyl decane, 6-Ethyl-3-octyl ester 2-trifluoromethyl benzoic acid |
10 BC, 11 HC |
100% sensitivity, 95% specificity |
Molecularly capped AuNPs—3 different ligands |
Chronic kidney disease |
healthy vs. stage 2: Isoprene, Acetone, Styrene, Toluene, 2-Butatone, 2,2,6-Trimethyl octane, 2,4-Dimethyl heptane Stage 2 vs. 3: Isoprene, Acetone, 2,2,6-Trimethyl octane, 2-Butatone, 2,4-Dimethyl heptane Stage 3 vs. 4: Acetone, Ethylene Glycol, Acetoin |
1–5 ppb |
42 CKD, 20 HC |
SVM |
79% accuracy early-stage CKD vs. HC 85% accuracy CKD stage 4 vs. stage 5 |
RT |
[4647] |
Molecularly capped AuNPs—5 different ligands |
Ovarian cancer |
Styrene, Nonanal, 2-Ethylhexanol, 3-Heptanone, Decanal, Hexadecane |
ppb level |
17 OV, 26 HC |
DFA |
82% accuracy |
RT |
[4445] |
Molecularly capped AuNPs—8 different ligands |
COVID-19 |
NA |
NA |
49 COVID-19, 33 non-COVID symptomatic, 58 HC |
DFA |
76% accuracy COVID-19 vs. HC 95% accuracy COVID-19 vs. non-COVID symptomatic |
RT |
[24] |
PAH-coated random SWCNTs network—4 different PAHs |
Multiple sclerosis |
Hexanal, 5-Methyl-undecane |
NA |
37 MS, 18 HC |
DFA |
80.4% accuracy |
RT |
[4950] |
Molecularly caped AuNPs/CDs-coated random SWCNTs network—20 different sensing films |
Alzheimer’s and Parkinson’s disease |
24 VOCs |
1–5 ppb |
15 AD, 30 PD, 12 HC |
DFA |
85% accuracy AD vs. HC 78% accuracy PD vs. HC 84% accuracy AD vs. PD |
RT |
[5051] |
Molecularly caped AuNPs/PAH-coated random SWCNTs network—20 different sensing films |
17 diseases (LC, CC, HNC, OC, BLC, PC, KC, GC, CD, UC, IBS, IPD, MS, PDISM, PH, PET, CKD) |
2-Ethylhexanol, 3-Methylhexane, 5-Ethyl-3-methyloctane, Acetone, Ethanol, Ethyl acetate, Ethyl benzene, Isononane, Isoprene, Nonanal, Styrene, Toluene, Undecane |
10 ppb |
813 any of 17 diseases, 591 HC |
DFA, HCA |
86% average accuracy |
RT |
[5152] |
Ligand capped Au, Pt, and CuNPs—6 different sensing films |
Human cutaneous leishmaniasis |
2,2,4-trimethyl pentane, 4-methyl-2-ethyl-1-pentanol, methyl vinyl ketone, nonane, 2,3,5-trimethyl hexane, hydroxy-2,4,6-trimethyl-5-(3-methyl-2 butenyl)cyclohexyl) methyl acetate, 3-ethyl-3-methyl heptane, octane, 2-methyl-6-methylene-octa-1,7-dien-3-ol |
NA |
28 HCL, 32 HC |
PCA, DFA |
98.2% accuracy, 96.4% sensitivity, 100% specificity |
RT |
[5253] |
pristine, COOH-, Hex-4T-Hex/DNA/oligomers, PTCDA/TAPC/TCTA monomers or PANI-functionalized SWCNTs |
COPD |
NH3, NO2, H2S, benzene, 2-propanol, acetone, ethanol, sodium hypochlorite, water |
sub-ppb |
12 COPD, 9 HC |
PCA |
Acetone, ethanol and 2-propanol selective PANI-, TAPC- and COOH-CNTs, respectively. NO2 relevant driver of real-samples classification. Larger clinical trials needed. |
RT |
[5354] |
Pristine WO3, 0.008 wt % |
| and 0.042 wt % Pt-WO3 macroporous NFs |
Halitosis |
H2S and Methyl mercaptan (in presence of Toluene and Acetone) |
sub-ppm |
4 simulated halitosis breath samples (1 ppm), 4 HC |
PCA |
Successful classification |
350 °C |
[5455] |
7 different commercial MOS |
Lung cancer |
Ethyl benzene, 4-Methyl octane, Undecane, 2,3,4-trimethyl hexane |
Down to a few ppb |
37 NSCLC (81.1% I, II), 48 HC |
PCA |
75% accuracy Promising prognostic tool after LC resection surgery |
300 °C |
[5556] |
5 different commercial MOS |
Lung cancer, COPD |
NA |
NA |
32 LC, 38 COPD, 72 HC |
PCA, SVM, k-nearest neighbors |
LC vs. HC: 91.3% accuracy, 84.4% sensitivity and 94.4% specificity COPD vs. HC: 90.9% accuracy, 81.6% sensitivity and 95.8% specificity |
NA |
[5657] |
Field Effect Transistor (FET)—arrays |
Molecularly modified SiNWs |
Gastric cancer |
2-Propenenitril, Furfural, 6-Methyl-5-heptene-2-one |
Down to a few ppb |
30 GC, 77 HC |
DFA |
>85% accuracy |
RT |
[5758] |
Molecularly modified SiNWs |
Gastric cancer |
2-Propenenitril, Furfural, 6-Methyl-5-heptene-2-one |
Down to a few ppb |
149 LC, 40 GC, 56 Asthma/COPD, 129 HC |
DFA, ANN |
>80% accuracy |
RT |
[4344] |
Lung cancer |
Heptane, Decane, 2-Methyl pentane, 2-Ethyl-1-hexanol, Propanal, Pentanal, Acetone |
Asthma/COPD |
Pentane |
Electrochemical sensor |
Commercial NO, CO sensors, carbon electrode with linear-aldehyde selective porous poly tetrafluoroethylene membrane |
Diabetes |
NO, CO, Formaldehyde, Acrolein, Propanal, Crotonaldehyde, Butanal, Pentanal, Hexanal, Heptanal, Octanal, Nonanal, Decanal, Acetaldehyde |
Low ppb |
15 diabetic, 14 HC |
|
LC vs. HC, diabetic vs. HC Cross-sensitivity for aldehyde sensor: Moderate for high level of ethanol and isopropanol/Weak for H2S, NO, methanol, 3-heptanone/None for NO2, propofol, isoprene, or acetone |
RT |
[5859] |
Lung cancer |
3 LC, 3 smokers, 3 HC |
|
Optical—Colorimetric sensor arrays |
24 chemically reactive colorants |
Lung cancer |
NA |
Low ppm |
92 LC, 137 HC |
LPM |
Accuracy 81.1% LC vs. HC, 82.5–89% one histology vs. HC, 86.4% ADC vs. SCC |
RT |
[5960] |
Optical sensors |
PMTFP-coated optical fiber |
Vit. E deficiency |
Ethane |
pmol/L |
20 HC |
NA |
NA |
RT |
[6061] |
Liver diseases, Schizophrenia, Breast cancer, Rheumatoid Arthritis |
Pentane |
Lung cancer |
Heptane, Octane, Decane, Benzene, Toluene, Styrene |
Piezoelectric (SAW) sensor arrays |
GC-column/Polyisobutylene-coated SAW, non-coated SAW sensors |
Lung cancer |
Styrene, Decane, Isoprene, Benzene, Undecane, 1-Hexene, Hexanal, Propyl benzene, Heptanal, 1,2,4-Trimethyl benzene, Methyl cyclopentane |
500 ppb |
20 LC, 15 HC, 7 chronic bronchitis |
ANN |
80% sensitivity and specificity |
RT |
[6][6162] |
Piezoelectric (QCM) sensor arrays |
7 different |
| metalloporphyrins |
COPD |
NA |
NA |
5 COPD per GOLD stage (20), 5 HC |
PLS-DA |
Fair repeatability of measurements within HC and hypoxemic COPD patients (stage 4) Potential COPD severity assessment |
RT |
[6263] |
8 different |
| metalloporphyrins |
Asthma |
NA |
NA |
27 asthma, 24 HC |
PCA, FNN |
87.5% accuracy |
RT |
[6364] |
8 different |
| metalloporphyrins |
Lung cancer |
NA |
NA |
20 LC, 10 HC |
PLS-DA |
85% accuracy LC vs. HC 75% accuracy ADC vs. SCC |
RT |
[6465] |
8 different |
| metalloporphyrins |
Lung cancer |
NA |
NA |
70 LC, 76 HC |
PLS-DA |
81% sensitivity, 100% specificity |
RT |
[6566] |
8 different |
| metalloporphyrins |
Tuberculosis |
NA |
NA |
51 TB (31/51 +HIV), 20 HC |
PCA, k-nearest neighbors |
94.1% sensitivity, 90% specificity |
RT |
[6667] |
7 different |
| metalloporphyrins |
Halitosis |
H2S, Butyric acid, Valeric acid |
10–15 ppb |
Oral malodor subjects, HC |
PCA |
PC1 78% of data variance |
50 °C |
[6768] |
8 different anthocyanins |
Asthma |
NA |
NA |
15 asthma, 27 HC |
Factor Analysis |
75% of total variance Repeatability similar to spirometry and eNO |
RT |
[6869] |
In vitro studies—Cell lines/Synthetic samples |
Chemiresistor arrays |
CNT-conductive polymer nanocomposites—5 different polymers |
Lung cancer |
Isopropanol, Tetrahydrofuran, Dichloromethane, Toluene, n-Heptane, Cyclohexane, Methanol, Ethanol, Water |
|
NA |
PCA |
High sensitivity and selectivity for all the analytes, PC1-PC3 98% of total variance, except the two alkanes |
RT |
[6970] |
Pristine rGO and rGO functionalized with 8 different amine ligands—9 elements |
Cancer |
Ethanol, 2-Ethylhexanol, Ethyl benzene, Nonanal |
25 ppm |
NA |
PCA |
Successful discrimination of VOCs The LOD and the effect of humidity have to be decreased |
RT |
[7071] |
Pristine Pd, ZnO and polypyrrole NWs |
Breast cancer |
Heptanal |
8.98 ppm |
NA |
PCA |
73.2% PC1 variance High sensitivity and specificity |
RT |
[7172] |
Acetophenone |
798 ppb |
2-Propanol |
129.5 ppm |
Isopropyl myristate |
134 ppm |
Pristine In2O3 and WO3 NRs, Au, Pt, or Pd NPs-decorated In2O3 and WO3 NRs—8 elements |
Diabetes |
Acetone |
1.48 ppb |
NA |
Polar plot |
Effective visual discrimination between the gases. Future PCA, DFA, HCA analysis. |
150–300 °C |
[7273] |
Asthma |
NO2 |
1.9 ppt |
Halitosis |
H2S |
2.47 ppb |
WO3 NTs |
| Pt NPs—WO3 NTs, |
| Pd NPs—WO3 NTs |
Asthma |
NO |
50 ppb |
NA |
NA |
NA |
350 °C |
[7374] |
Lung cancer |
Toluene |
100 ppb |
400 °C |
Pristine, 0.1% wt GO- and 0.1 wt % thin layered graphite WO3 Hemi tubes |
Diabetes |
Acetone |
1 ppm |
NA |
NA |
NA |
350 °C |
[7475] |
Halitosis |
H2S |
Electrochemical sensor |
MWCNTs/Au-Ag NPs/GCE |
Gastric cancer |
3-Octanone |
0.3 ppb |
MGC-803 GC and GES-1 gastric mucosa cell lines |
NA |
Easy cell line discrimination, high sensitivity, good VOCs selectivity in presence of CO2, acetone and ethanol |
RT |
[7576] |
Butanone |
0.5 ppb |
SiNWs-rGO |
Infectious diseases |
Cyclohexane, Formaldehyde in presence of Methanol, Ethanol, Acetonitrile, Acetaldehyde and humidity |
1 ppm |
NA |
|
Novel electrode platform with increased sensitivity, selectivity and repeatability |
|
[7677] |
Piezoelectric (SAW) sensor arrays |
SH-Calix4arene, |
| AuNRs, AgNCs, |
| Calix4arene-AuNRs, Calix4arene-AgNCs |
Lung cancer |
Chloroform, Toluene, Isoprene, Acetone, n-Hexane, Ethanol |
1.52–12.34 ppm for CHCl3 1.54–2.64 ppm for toluene |
NA |
NA |
Sensitivity ↑ for all VOCs Chloroform, toluene: 6–8 times higher sensitivity than individual responses Selectivity ↑: modified AuNRs for CHCl3, modified AgNCs for Toluene |
RT |
[7778] |
Pristine or AuNPs-functionalized zeolitic-imidazole-framework nanocrystals (ZIF-8, ZIF-67) |
Diabetes |
Acetone, Ammonia, Ethanol |
acetone 1.1–3.6 ppm, ethanol 0.5–3 ppm NH3 1.6–3.2 ppm |
NA |
PCA |
Effective discrimination of diabetes biomarkers |
RT |
[7879] |
Piezoelectric (QCM) sensor arrays |
TiO2-MWCNTS and Cobalt (II) phthalocyanine-silica on Au layers |
Diabetes |
Acetone |
4.33 ppm |
NA |
NA |
High sensitivity |
RT |
[7980] |
Asthma |
NO |
5.75 ppb |
Optical—Colorimetric arrays |
36 chemically responsive dyes (porphyrin derivatives, NaFluo) |
Lung cancer |
p-Xylene, Styrene, Isoprene, Hexanal |
50 ppb |
NA |
HCA, PCA, BPNN |
100% accuracy of kind and concentration discrimination, promising for real-sample experiments |
RT |
[8081] |
AuNRs-modified metalloporphyrins and pH responsive dyes—36 spots |
Lung cancer |
Decane, Undecane, Hexanal, Heptanal, 1,2,4-Trimethylbenzene, Benzene |
<1 ppm |
NA |
PCA, HCA |
64.2% accuracy of structurally similar VOCs, 93% photoprotection of metalloporphyrins, ↑ repeatability and long-term stability |
RT |
[8182] |
AD: Alzheimer’s Disease, ADC: Adenocarcinoma, ANN: Artificial Neural Network, BC: Breast cancer, BLC: Bladder cancer, BPNN: Back-Propagation Neural Network, BUN: Blood urea nitrogen, CC: Colorectal cancer, CD: Crohn’s Disease, CDs: Cyclodextrin derivatives, CKD: Chronic Kidney Disease, COPD: Chronic Obstructive Pulmonary Disease, DFA: Discriminant Function Analysis, FNN: Feet-forward Neural Network, GC: Gastric cancer, GCE: Glass Carbon Electrode, GOLD: Global Initiative for Obstructive Lung Disease, HC: Healthy control, HCA: Hierarchical Cluster Analysis, HNC: Head and Neck cancer, IBS: Irritable bowel syndrome, IPD: Idiopathic Parkinson’s disease, KC: Kidney cancer, LC: Lung cancer, LOD: Limit of Detection, LPM: Logistic prediction model, MLP: Multi-layer Perceptron, MS: Multiple Sclerosis, MWCNTs: Multi-wall carbon nanotubes, NA: Not Applicable, NaFluo: sodium fluorescein, NCs: Nanocubes, NFs: Nanofibers, NPs: Nanoparticles, NRs: Nanorods, NSCLC: Non-small Cell Lung Carcinoma, OC: Ovarian cancer, PAH: Polycyclic aromatic hydrocarbons, PC: Prostate cancer, PCA: Principal Component Analysis, PD: Parkinson’s Disease, PDISM: Atypical Parkinsonism, PET: Pre-eclampsia toxemia, PH: Pulmonary Hypertension, PLS-DA: Partial Least Square Discriminant Analysis, PMTFP: Polymethyl(3,3,3-trifluoropropyl)siloxane, RT: Room temperature, SCC: Squamous cell carcinoma, SiNWs: Silicon nanowires, SVM: Support Vector Machine, T: Temperature, TB: Tuberculosis, TFB: Poly(9,9-dioctylfluorenyl-2,7-diyl)-co-(4,4′-(N-(4-s-butylphenyl)diphenylamine), UC: Ulcerative Colitis.
Table 3. Sensing devices used for differential diagnosis, staging, and phenotyping of different categories of diseases.
Sensor |
Diseases/Phenotypes/Stages |
Subjects |
Classifier |
Results |
Ref. |
Differential diagnosis |
Cyranose 320 |
NSCLC vs. COPD (GOLD stage I-III) |
10 NSCLC, 10 COPD |
PCA, CDA |
85% acc. |
[8283] |
LC vs. non-cancer (COPD, asthma, pneumonia, pulmonary embolism, BPN) |
165 LC, 91 non-cancer |
SVM |
87.3% acc. |
[8384] |
LC vs. COPD vs. LC/COPD vs. HC |
63 LC, 15 COPD, 79 both, 78 HC |
77.4% acc., 100% accurate LC/COPD classification |
LC vs. non-cancer (COPD, asthma, pneumonia, pulmonary embolism, bronchiectasis, BPN, TB) |
252 LC, 223 non-cancer |
LRA |
Sens.: 95.8% (S), 96.2% (NS) Spec.: 92.3% (S) 90.6% (NS) |
[8485] |
Asthma vs. COPD |
20 asthma, 30 COPD |
PCA, CDA |
96% acc. Within/between day repeatability, reproducibility of e-Noses |
[8586] |
Fixed and classic asthma vs. COPD (GOLD stages II-III) |
21 fixed asthma, 40 COPD |
PCA, CDA |
88% acc., 85% sens., 90% spec. |
[8687] |
39 classic asthma, 40 COPD |
83% acc., 91% sens., 90% spec. |
IPF vs. COPD |
32 IPF, 33 COPD |
PCA, CDA |
80% cross-validated acc., Wider patient cohorts and inclusion of more comorbidities needed |
[8788] |
COPD vs. LC vs. BC |
50 COPD, 30 LC, 50 BC |
PCA, CDA, CAP |
Correct classification values: LC vs. COPD 96.47%, LC vs. BC 93.05%, BC vs. COPD 100%, COPD vs. LC vs. BC 91.35% |
[4849] |
Bronchial vs. Laryngeal SCC (advanced) |
10 bronchial, 10 laryngeal |
JMP Pro |
10% misclassification, 100% sens., 80% spec. |
[8889] |
AD vs. PD vs. HC |
18 AD, 16 PD, 19 HC |
LDA |
76.2% sens., 45.8% spec., p < 0.0001 |
[8990] |
AeoNose |
ILDs (COP, CTD) vs. COPD, ILDs subgroups (COP, HP, IPF, sarcoidosis, uILD, asbestosis, NSIP, RB-ILD, DIP) |
28 COP, 23 COPD |
Athena program, t-test |
AUC 0.77, 75% sens., 71% spec. |
[9091] |
25 CTD-ILD, 23 COPD |
AUC 0.85, 88% sens., 71% spec. |
174 ILDs, 23 COPD |
Less accurate discrimination of ILDs subgroups (e.g., AUC IPF vs. CTD-ILD 0.86, COP vs. CTD-ILD 0.82) |
Asthma vs. CF |
20 asthma (moderate-severe), 13 CF |
ANN |
AUC 0.90, 89% sens., 77% spec. |
[9192] |
HNSCC vs. LC |
52 HNSCC, 32 LC |
ANN |
Acc., sens., spec.: 93%, 96%, 88% (best fit), 85%, 85%, 84% (cross-validation) |
[9293] |
Cancer types |
100 HNSCC, 40 bladder, 28 colon cancer |
ANN |
Acc., sens., spec.: 81%, 79%, 81% HNSCC vs. colon cancer, 84%, 80%, 86% HNSCC vs. bladder cancer, 84%, 88%, 79% Colon vs. bladder cancer |
[9394] |
SpiroNose |
LC vs. COPD vs. asthma vs. HC |
31 LC, 31 COPD, 37 asthma, 45 HC |
PCA |
Cross-validation values 78–88%, repeatability ↑ |
[9495] |
ILD subgroups: |
141 sarcoidosis, 85 IPF, 33 CTD-ILD, 25 HP, 11 IPAF, 10 NSIP |
PLS-DA |
Acc., sens., spec.: 77%, 75%, 84% IPF vs. HP, 94%, 98%, 85% IPF vs. CTD-ILD, 92%, 92%, 90% IPF vs. NSIP, 89%, 87%, 100% IPF vs. IPAF, 75%, 100%, 67% CTD-ILD vs. IPAF, 98%, 90%, 100% CTC-ILD vs. NSIP, 90%, 94%, 72% HP vs. sarcoidosis, 91%, 92%, 88% (training), 91%, 95%, 79% (validation) IPF vs. non-IPF |
[9596] |
Chemiresistor-based |
| alkane sensor |
LC vs. HC LC vs. COPD |
12 LC, 12 COPD, 13 HC |
MANOVA |
LC: 83.3% sens., 88% spec. Sensor acc no smoke-dependence |
[9697] |
MOS, electrochemical, hot wire, and catalytic |
LC vs. COPD |
48 LC, 52 COPD |
8 different |
76.9–84.75% acc., 75–81.36% sens., 78.79–88.14 spec. Highest acc. With KPCA-XGBoost |
[9798] |
LC vs. COPD |
33 LC, 28 COPD |
PCA-SVM, KPCA-SVM, PCA-XGBoost, KPCA-XGBoost |
82.52–96% acc., 78.33–95% sens., 85–96.67% spec. Highest acc. With KPCA-XGBoost |
[9899] |
Organically-coated AuNPs and SWCNTs based chemiresistor |
LC (I/II) vs. BPN |
16 LC, 30 BPN |
DFA |
87% acc., 75% sens., 93% spec. Low LC sample → careful interpretation |
[99100] |
BC vs. benign |
30 HC, 15 BBT, 13 DCIS, 96 BC |
DFA |
Acc., sens., spec.: 88.3%, 90.6%, 83.3% BC vs. BBT/HC, 71.2–82%, 62.6–80%, 75.7–82.3% BC vs. BBT, 81.4–84.4%, 83–83.3%, 81–92% BC vs. DCIS |
[100101] |
Gca vs. OLGIM groups (0-IV) |
99 Gca, 155 OLGIM 0, 136 OLGIM I-II, 34 OLGIM III-IV, 53 PUD |
DFA |
Acc., sens., spec.: 92%, 73%, 98% Gca vs. 0–IV, 84%, 90%, 80% Gca vs. 0, 87%, 97%, 84% Gca vs. 0–II, 90%, 93%, 80% Gca vs. III-IV, 85%, 93%, 80% Gca vs. I–IV, 87%, 87%, 87% Gca vs. PUD |
[101102] |
Gca vs. benign gastric conditions |
37 Gca, 32 ulcers, 61 less severe conditions |
DFA |
89% sens., 90% spec. 84% sens., 87% spec. |
[102103] |
ulcer vs. less severe |
AD vs. PD AD vs. PD vs. HC |
15 AD, 30 PD, 12 HC |
DFA |
AD vs. PD: 84% acc., 80% sens., 87% spec. Feasible overall discrimination, with large PD/HC overlap |
[5051] |
NA-NOSE |
BC, benign breast conditions, normal mammographs |
11 BC, 14 benign, 7 normal mammographs |
PCA/ANOVA/Student’s t-test, SVM |
94% sens., 80% spec. for benign vs. BC and negative mammography, Similar results with both methods |
[103104] |
MCNPs-based chemiresistor—6 sensors-array |
IBD vs. IBS |
71 IBD (35 UC, 36 CD), 26 IBS |
ANN |
81/88% acc., 92/73% sens., 53/100% spec. (real/artificial) |
[104105] |
CD vs. UC |
75/96% acc., 75/100% sens., 47/93% spec. (real/artificial) |
Molecularly modified SiNW FET |
Gca vs. LC |
40 Gca, 149 LC, 56 asthma/COPD |
DFA, ANN |
92% acc., 93% LC, and 85% Gca correct classification |
[4344] |
LC vs. asthma and COPD |
89% acc., 92% sens., 80% spec. |
MOS gas sensor array |
Gca vs. gastric ulcer patients |
49 Gca, 30 gastric ulcer |
Back-propagation Neural network |
93% acc., 94.38% sens., 89.93% spec. Classification acc. Of malignant, benign, normal subjects: 92.54%, 93.17%, 92.49%. |
[105106] |
OC vs. benign and HC |
86 OC, 51 benign, 114 HC |
PCA, k-NN |
Acc., sens., spec.: 85%, 6%, 84% (cross-validation/strict prediction), 87%, 89%, 86% (prediction/strict prediction), 86%, 84%, 85% (cross-validation/most probable pred.), 100%, 100%, 100% (prediction/most probable pred.) |
[106107] |
AD vs. PD vs. HC |
20 AD, 20 PD, 20 HC |
PCA |
Effective discrimination of AD vs. PD and HC |
[107108] |
BIONOTE |
CLD vs. NC-CLD |
65 CLD, 39 NC-CLD |
PLS-DA, radar plot |
Successful discrimination, 16 cirrhotic patients misclassified |
[108109] |
Commercial (MQ) gas sensors |
CKD vs. diabetes vs. HC high creatinine, HC low creatinine |
84 CKD, 24 diabetes, 54 HC high creatinine, 54 HC low creatinine |
Radar plots, PCA, SVM, PLS-regression, HCA |
PCA: 96.64% of total variance expressed in PC1–3 SVM: 100% correct classification of samples |
[109110] |
Disease histology/phenotyping |
Tor Vergata e-Nose |
SCC vs. ADC |
10 SCC, 10 ADC |
PLS-DA |
75% correct classification |
[6465] |
24 colorants |
SCC vs. ADC |
22 SCC, 50 ADC |
LPM |
86.4% acc. |
[5960] |
SCLC vs. NSCLC |
9 SCLC, 83 NSCLC |
78.1% acc. (moderate) |
UV-irradiated pristine, Au, Pt, Au/Pt, Ni, Fe-doped WO3NWs |
LC vs. HC |
4 SCLC, 8 SCC, 10 ADC, 12 HC |
PCA |
98.6 % acc. |
[110111] |
SCLC vs. NSCLC, SCC vs. ADC |
DFA |
Acc.: 84.5% SCLC vs. NSCLC,77.5% SCC vs. ADC |
Molecularly capped AuNPs and SWCNT based chemiresistors |
LC with vs. without EGFR mutation |
19 with EGFR, 34 without EGFR |
DFA |
83% acc., 79% sens., 85% spec. |
[99100] |
Cyranose 320, Tor Vergata, Common In-vent, Owlstone Lonestar |
Clinically stable vs. unstable episodes of asthma |
22 partly controlled persistent asthma |
PCA |
Correct classification: 95% baseline vs. loss of control, 86% loss of control vs. recovery Owlstone Lonestar the most prominent |
[111112] |
Cyranose 320 |
Asthma inflammatory phenotypes |
24 EOS., 10 NEUTR., 18 PAUC. |
PCA |
Acc., sens., spec.: 73%, 60%, 79% EOS. vs. NEUTR., 74%, 55%, 87% EOS vs. PAUC., 89%, 94%, 80% NEUTR. vs. PAUC. |
[112113] |
Uncontrolled asthma-like symptoms |
Training set: 65 cluster 1, 22 cluster 2, 34 cluster 3 |
one-way ANOVA, Kruskal-Wallis |
Significant differences concerning chest tightness during exercise, dyspnea and gender |
[113114] |
HC and controlled vs. partly controlled and uncontrolled asthma |
10 HC, 9 controlled, 7 partly, 12 uncontrolled |
PCA, radar plot |
Good predictive ability Cross-validated AUC 0.80, 79% sens., 84% spec. |
[114115] |
Organically-coated AuNPs and SWCNT-based chemiresistor |
BC subtypes |
12 LuminalA, 42 LuminalB, 12 Triple Negative, 16 HER2+, 14 HER2 equivocal |
DFA |
Acc., sens., spec.: LuminalA vs. others 81.3–87.7%, 75–87.5%, 82.1–87.5% LuminalB vs. others78.1–86.3%, 83.3–85.3%, 74.1–87.2% HER2+ vs. others 81.3–82.4%, 81.3–91%, 80.7–81.3% Triple neg, vs. others 82.9–90.3%, 83.3–93.3%, 82.9–89.4% Luminal vs. non-Luminal 70.8–87.7%, 70.4–88.1%, 71.4–87.1% LuminalA vs. LuminalB 85.7–94%, 75–91.7%, 88.2–95.2% HER2 status/luminal 85.7–100%, 85.7–100%, 83.3–100% HER2 status/non-luminal 90.9%, 90.9%, 90.9% |
[100101] |
Disease staging |
Tor Vergata e-Nose |
LC Stage I vs. II/III/IV |
40 stage I, 18 stage II, 6 III/IV |
PLS-DA |
Sens.: stage I 90% vs. stage II-IV 57% (+ metabolic diseases), stage I 96% vs. stage II-IV 60% (LC only) |
[6566] |
24 colorants |
LC Stage I/II vs. LC stage III/IV |
41 SCLC, 42 NSCLC |
LPM |
79.3 % acc. (moderate) |
[5960] |
11 sensor-array (MOS, electrochemical, hot wire and catalytic) |
LC Stage III vs. IV |
44 stage II, 46 stage IV |
PCA-SVM, KPCA-SVM, PCA-XGBoost, KPCA-XGBoost |
70.42–82.42% acc., 45–81% sens., 79–95.5% spec. |
[9899] |
Organically-coated AuNPs and SWCNTs based chemiresistor |
OLGIM stages |
155 OLGIM 0, 136 OLGIM I-II, 34 OLGIM III-IV, 7 Dysplasia |
DFA |
Acc., sens., spec.: 0-II vs. III-IV and dysplasia 61%, 83%, 60%, 0 vs. I-II 43%, 45%, 41%, 0 vs. III-IV 66%, 90%, 61%, 0 vs. I-IV 50%, 50%, 50%, I-II vs. III-IV 64%, 80%, 60% |
[101102] |
GCa I-II vs. III-IV |
17 GCa I-II, 18 GCa III-IV |
DFA |
89% sens., 94% spec. |
[102103] |
Molecularly modified SiNW FET |
LC staging (I-II vs. III-IV) |
34 early stage, 110 advanced stage |
DFA, ANN |
81% acc., 34.5%sens.,95% spec. |
[4344] |
GCa staging (I-II vs. III-IV) |
86.5% correct classification, 84.6 early stage, 87.5 advanced |
Cyranose 320 |
Bronchial/Laryngeal in situ vs. advanced |
bronchial: 10 in situ, 10 advanced, laryngeal: 12 in situ, 10 advanced |
JMP Pro |
21% misclassification rate, 82% sens., 75% spec. |
[8889] |
BIONOTE |
Liver cirrhosis (A, B, C Child–Pugh) |
NA |
PLS-DA |
Successful discrimination |
[108109] |
Acc.: accuracy, AD: Alzheimer’s disease, ADC: adenocarcinoma, AUC: Area Under the Receiver Operating Characteristic Curve, ANN: artificial neural network, BC: breast cancer, BBT: Breast benign tumor, BPN: benign pulmonary nodules, CAP: canonical analysis of principal coordinates, CD: Crohn’s Disease, CDA: canonical discriminant analysis, CF: cystic fibrosis, CLD: chronic liver disease, COP: cryptogenic organizing pneumonia, COPD: chronic obstructive pulmonary disease, CTD-ILD: connective-tissue diseases-associated ILD, DCIS: ductal carcinoma in situ, DFA: Discriminant Function Analysis, DIP: desquamative interstitial pneumonia, EOS: eosinophilic, GCa: gastric cancer, GOLD: Global Initiative for Obstructive Lung Disease, HBC: hexabenzocoronene, HCA: Hierarchical Cluster Analysis, HER2: Human epidermal growth factor receptor 2, HNSCC: Head and neck squamous cell carcinoma, HP: hypersensitivity pneumonitis, IBD: Inflammatory Bowel Diseases, IBS: Irritable Bowel Syndrome, ILDs: interstitial lung diseases, IPAF: interstitial pneumonia with autoimmune features, IPF: idiopathic pulmonary fibrosis, k-NN: k-Nearest Neighbors, KPCA-XGBoost: kernel principal component analysis—extreme gradient boosting, LC: Lung cancer, LRA: logistic regression analysis, LPM: Logistic prediction model, MANOVA: multivariate analysis of variance, MOS: metal oxide semiconductor, NC-CLD: non-cirrhotic CLD, NEUTR: neutrophilic, NS: non-smokers, NSCLC: non-small cell lung cancer, NSIP: non-specific interstitial pneumonia, NWs: nanowires, OC: ovarian cancer, OLGIM: operative link on gastric intestinal metaplasia, PAH: Polycyclic aromatic hydrocarbons, PAUC: paucigranulocytic, PCA: Principal Component Analysis, PD: Parkinson’s disease, PLS-DA: Partial Least Square Discriminant Analysis, PUD: peptic ulcer disease, RB-ILD: respiratory bronchiolitis-associated ILD, S: smokers, SCC: squamous cell carcinoma, SCLC: small cell lung cancer, Sens.: sensitivity, Spec.: specificity, SVM: Support Vector Machine, TB: Tuberculosis, UC: Ulcerative Colitis, uILD: unclassifiable ILD, UV: ultra-violet.
2.1. The Case for Lung Diseases
Chronic and acute lung diseases such as asthma, COPD, idiopathic pulmonary fibrosis (IPF), LC, mesothelioma, and sarcoidosis have been connected with similar metabolic alternations
[115116]. Especially asthma and COPD are also characterized by similar symptoms
[37] with COPD being commonly underdiagnosed or diagnosed at late stages
[116117]. Concerning LC, no symptoms are expressed in early stages
[117118] while disease manifestation is limited to non-specific symptoms
[117118] including cough, short breath, chest pain, and weight loss
[37]. Disease phenotyping, on the other hand, is mandatory in some cases. Asthma subtypes such as eosinophilic, neutrophilic, mixed granulocytic, and paucigranulocytic asthma
[118119] are characterized by similar symptoms while different treatment is required
[119120]. Similarly, immunosuppressive, antifibrotic, or a combination of medications may be needed for fibrotic interstitial lung diseases (ILDs), depending on the respective phenotype (inflammatory, more fibrotic, or combination)
[9596]. Thus, reliable phenotyping is needed for appropriate medication to be administered
[33][119120][9596]. LC is subdivided into different categories with different clinical characteristics as well. Small cell LC (SCLC), with 20–25% percentage of occurrence
[120121], is characterized by increased metabolic and proliferation rates compared to other cancer cells
[121122], while non-small cell LC (NSCLC) accounts for 70–75% of LC cases and is subdivided into the smoking-related
[37][122123] squamous cell carcinoma (SCC)
[120121] and the non-squamous cell carcinomas
[37] including adenocarcinomas (ADC) (minor smoking correlation) and large-cell carcinoma (LCC)
[120121]. Consequently, the accurate discrimination of different lung diseases and subtypes of a lung disease, especially using breath analysis of exhaled VOCs, is of particular importance.
The use of GC-MS has rendered lung disease differentiation, phenotyping and staging feasible in many cases. To be more specific, LC discrimination from patients with other lung diseases has been investigated in several studies. In an attempt to discriminate between NSCLC, COPD, and HC patients, by taking smoking habits into account, 4 VOCs were identified (in varying concentrations) for NSCLC and COPD
[123124]. In another study, Wang et al. attempted to discriminate LC from COPD, asthma, pneumonia, pulmonary embolism and benign lung tumor patients; however, the selected 10 VOCs could not discriminate accurately between the two groups, implying their potential confounding role during LC-biomarkers determination
[18]. Koureas et al. have also attempted to discriminate LC from other respiratory diseases, using 19 distinctive VOCs, based on the underlying disease mechanisms (targeted method); only the discrimination of LC patients from HC, using ethylbenzene, toluene, styrene, 2- and 1-propanol was achieved
[4748]. However, in a more recent study of the same group, the discrimination of LC patients from patients suffering from sarcoidosis, hypersensitivity pneumonitis, interstitial lung diseases or pulmonary infections was achieved with an increased accuracy of 75.3%; the 29 VOCs were selected following a hypothesis-generating non-targeted strategy
[124125]. In a different study, LC was accurately distinguished from pulmonary non-malignant diseases (PNMD; COPD, pulmonary tuberculosis, asthma) using 10 VOCs while 5 were selected as characteristic of LC in contrast to both PNMD and HC
[125126].
The differentiation of LC patients from patients with benign pulmonary nodules (BPN) has also been extensively reported. Apart from a study by Wang et al.
[18] which included patients with benign lung tumors, Fu et al. have investigated the respective discriminant ability of carbonyl VOCs
[126][127][128][129]. Four carbonyl VOCs, captured by a silicon micro-reactor, were found to present increased concentration in LC patients when compared to BPN patients and HC
[126127]. In a subsequent study, the same group achieved the differentiation of both early and III, IV stage LC patients from BPN patients, with high sensitivity (83%) and particularly increased specificity (74%) in comparison to positron emission tomography (90% and 39%, respectively)
[127128], while 6 carbonyl VOCs have permitted a classification accuracy of 89% of LC vs. BPN patients
[128129]. More recently, Chen et al. identified 19 VOCs able to distinguish not only LC and BPN patients (this with an accuracy of 80.9%) but also early-stage LC patients from BPN (with an accuracy of 75.6%), being remarkably promising for early LC diagnosis
[117118].
LC histology and staging characterization using analytical methods is another important target of this research field. It was reported that 1-butanol, 3-hydroxy-2-butanone
[9], as well as 4-hydroxyhexenal
[126127], can differentiate SCC from ADC patients with the former being decreased for SCC in contrast to the other 2 VOCs. Similarly, SCLC and NSCLC can be potentially distinguished from 4-hydroxynonenal and C5H10O
[126127]. Hexanal has also been found in higher concentrations for SCLC patients compared to NSCLC, potentially due to increased metabolic rates
[121122]; Chen et al. have achieved NSCLC and SCLC differentiation, with an accuracy of 93.9%, using a pattern of 20 VOCs
[117118]. Concerning LC staging (I, II, III, or IV), a pattern of 19 VOCs was used to distinguish between early (I, II) and advanced LC stages (III, IV) with 82.7% accuracy
[117118] while Fu et al. demonstrated that exhaled 2-butanone concentration is significantly different between stages I and II–IV
[126127].
Apart from LC, other lung diseases are also studied for accurate diagnosis. A series of studies have focused on asthma phenotyping using GC-MS. Brinkman et al. identified 3 VOCs significantly correlated with sputum eosinophils
[111112], while Ibrahim et al. identified VOC-patterns differentiating eosinophilic from non-eosinophilic (6 VOCs) and neutrophilic from non-neutrophilic (7 VOCs)
[129130]. Recently, Schleich et al. identified 4 VOCs discriminating eosinophilic from neutrophilic, eosinophilic from paucigranulocytic and neutrophilic from paucigranulocytic asthma, with accuracy similar to blood eosinophils and FeNO tests
[118119]. In a more recent study, the same group used two-dimensional GC-high resolution-time-of-flight-MS, selected ion flow tube mass spectrometry (SIFT-MS), 10 VOCs, and 9 ion channels so as to achieve asthma phenotyping with an accuracy of 75%
[119120]. COPD phenotyping and staging has also been attempted using analytical techniques. Fens et al. identified 8 eosinophils- and 17 neutrophils-related VOCs, with only one VOC overlapping between the two subgroups. More VOCs were related with cell counts for Global Initiative for Obstructive Lung Disease (GOLD) stage II, in comparison to GOLD stage I
[130131]. In another study, 11 COPD patients with >1% and 6 with >2% eosinophil count were discriminated from non-eosinophilics (<1% and <2% eosinophil count, respectively) with accuracies of 79% and 92%
[131132]. Exacerbation prediction of both asthmatic children
[132][133][134][135] and adults
[111112][129130] as well as COPD patients
[135136] also comprises a subject of study.
Following the promising applications of analytical methods that highlight the potential capabilities of breath analysis in phenotyping, staging and differential diagnosis of lung diseases, sensing devices have also been used in respective applications with remarkable results. Research interest has focused on LC discrimination from other lung diseases such as COPD and asthma. Cyranose 320 has achieved separation of NSCLC from COPD (GOLD stage I–III) with an accuracy of 85%, in an article by Dragonieri et al.
[8283]. Tirzīte et al. have used this e-Nose to effectively discriminate, not only LC patients from COPD, asthma, pneumonia, pulmonary embolism, benign lung tumor patients, and HC, with 87.3% accuracy, but also between LC patients, COPD patients, LC patients suffering also from COPD and HC with 77.4% accuracy, and totally correct classification of the 79 LC/COPD patients
[8384]. In a more recent study, the same group discriminated LC from patients with non-malignant lung diseases as well as bronchiectasis, tuberculosis, and HC by taking into account smoking habits. An overall sensitivity and specificity of 95.8% and 92.3% for smokers and 96.2% and 90.6% for non-smokers, respectively, was observed using Cyranose 320
[8485]. More recently, the same e-Nose was used by Rodriguez et al. for the discrimination of COPD from LC and BC, achieving an overall correct classification of 91.35% while LC correct classification in relation to COPD was equal to 96.47%
[4849]. Interestingly, the contribution of the 32 sensors in the discrimination was also assessed
[4849]. Tor Vergata e-Nose has been used effectively for discriminating LC patients from COPD, Interstitial lung disease, Pleurisy and Bronchitis patients, with a sensitivity of 89.3% for LC patients
[136137]. In a particularly promising study, SpiroNose discriminated LC, asthma, COPD, and HC, with the respective accuracy values presented in
Figure 4a (68–88%)
[9495]. The applicability of breath sampling and analysis was tested as the collection of asthma breath samples at two different sites led to similar results (
Figure 4b)
[9495]. In the field of non-commercial and self-developed sensors, Tan et al. have attempted to develop a cross-reactive alkane-based chemiresistor combining carbon powder and tetracosane, achieving not only high affinity for alkanes and low sensitivity for polar VOCs (water, ethanol, ethanal) but also effective differentiation of 12 LC patient from 13 HC and 12 COPD patients
[9697]. In a more recent study, researchers attempted to differentiate LC, COPD, and asthma patients from HC, using an array of 8 sensors of 4 different types (MOS, electrochemical, hot wire, and catalytic type). The array achieved accuracy between 76.9–84.75%, using different machine learning methods
[9798]. Accuracy values were greater for LC and COPD prediction; however, the maximum accuracy value of 84.75% was attained using kernel principal component analysis—extreme gradient boosting (KPCA-XGBoost), which indicates excellent discriminatory capability for LC and COPD patients
[9798]. Similarly, using an array of 11 sensors of 4 different types (namely MOS, electrochemical, hot wire, and catalytic type), Liu et al. differentiated non-smoking LC and COPD patients, with the best discriminatory accuracy (96%) being achieved using the same machine learning technique
[9899]. The discrimination of LC from asthma and COPD patients was also achieved by Haick’s group with particularly high classification accuracies
[4344]. Early-stage LC discrimination from BPN has been reported by Haick et al. using an array of 40 chemiresistors based on MCNPs (Au NPs) and molecularly-coated SWCNTs, achieving an accuracy of 87%. Considering that the required treatment may change in the occurrence of genetic alternations, the differentiation of patients with and without epidermal growth factor receptor (EGFR) mutation was also attempted with an accuracy of 83%
[99100].
Figure 4. (
a) Presentation of the cross-validation percentages of the differentiation of asthma, COPD, LC patients, and HC, using SpiroNose; (
b) PCA plot of breathprints collected from asthmatic patients at the Academic Medical Center (AMC), Amsterdam and Medical Spectrum Twente (MST), Enschede, for which no significant differentiation is observed (p = 0.892). Adapted with permission from Ref.
[9495]. Copyright © 2015 IOP Publishing Ltd.
Concerning LC histology and staging with sensing devices, promising studies have been reported in the literature. The discrimination of NSCLC subtypes ADC and SCC has been permitted using Tor Vergata e-Nose with an accuracy of 75%, by applying endoscopic breath sampling
[6465] as well as by using a colorimetric sensor-array of 24 elements developed by Mazzone et al. ultimately achieving an accuracy of 86.4%
[5960]. SCLC and NSCLC differentiation and LC staging (I/II vs. III/IV) were also examined by Mazzone et al. though with moderate accuracies
[5960]. A 6-sensor-array based on UV-irradiated (394 nm) pristine or metal-doped WO3NWs (
Table 3) differentiated effectively not only ADC from SCC, but also between SCLC and NSCLC with 77.5% and 84.5% accuracy values, respectively (
Figure 5)
[110111]. In another study aiming at the discrimination of LC patients from HC while taking into account the existence of metabolic comorbidities, Tor Vergata e-Nose exhibited far higher sensitivity for stage I LC in comparison to the rest of stages, either in the presence or absence of metabolic diseases (
Table 3)
[6566]. LC staging was recently attempted by Liu et al. along with COPD discrimination as mentioned above, with stage III LC being effectively discriminated from stage IV with an accuracy higher than 80%, using KPCA-XGBoost
[9899]. Haick’s team has achieved LC staging with an accuracy of 81% and with low sensitivity, using a molecularly modified Si NW FET (
Table 3)
[4344].
Figure 5. DFA plots representing the discrimination of (
a) LC patients from HC; (
b) SCLC from NSCLC patients; and (
c) SCC from ADC patients, using a 6-sensor array of UV-irradiated (394 nm) pristine or metal-doped WO3NWs. The arrays achieved the detection of lung cancer but also the prediction of LC histological subtypes. Reprinted with permission from Ref.
[110111]. Copyright © 2020 Published by Elsevier B.V.
As in the case of breath analysis with analytical methods, precise diagnosis of lung diseases other than LC via sensing devices is an extensive field of research. The effective discrimination of COPD and asthma has been reported in the literature by Fens at al. using Cyranose 320 and taking into consideration smoking habits, leading to high cross-validated accuracy values (
Table 3)
[8586][8687]. More recently, asthma and CF discrimination was also reported for pediatric population using AeoNose and with high accuracy values, excluding the confounding factors of diet, exercise, comorbidities and inhaled drugs
[9192]. Concerning ILDs, Krauss et al. used the AeoNose in an attempt to differentiate between ILDs subgroups (
Table 3), with moderate accuracy, as well as between ILDs cryptogenic organizing pneumonia and connective-tissue diseases-associated ILD from COPD patients with good sensitivity and specificity
[9091]. COPD and IPF differentiation has been recently investigated by Dragonieri et al., with a high accuracy of 80%, verified by external validation using Cyranose 320
[8788]. In contrast to Krauss et al., Moor’s group achieved to reliably discriminate patients suffering from different ILDs by using SpiroNose as well as greater cohorts of ILD-patients (
Table 3)
[9596]; the group demonstrated the applicability of e-Noses in ILDs differential diagnosis and specifically in IPF discrimination from non-IPF patients with high accuracies (91%)
[9596].
Disease phenotyping using sensing devices seems to be also feasible. Plaza et al. achieved differentiation between the three inflammatory phenotypes of asthma with high accuracy values (
Table 3), with the participants’ phenotypes being characterized by differential leukocyte counts in induced sputum
[112113] while asthma-control assessment has been also reported. Brinkman et al. used 4 different e-Noses in order to discriminate between stable and unstable periods, comparing baseline (control) vs. loss of control and loss of control vs. recovery breath samples, with the Owlstone Lonestar being the most prominent concerning the discrimination of unstable periods (
Table 3)
[111112]. More recently, Moreira et al. demonstrated the ability of Cyranose 320 to discriminate the uncontrolled asthma-like symptoms, using 3 different groups of asthmatic or suspicious of asthma participants divided by unsupervised hierarchical clustering
[113114]. The division of participants was based on asthma, lung function, symptoms of the last month, age, and food/drink intake 2 h before breath sampling
[113114]. In another recent study, the same e-Nose was used for the effective discrimination of HC and asymptomatic-controlled asthmatic children from the symptomatic partly-controlled and uncontrolled asthmatic children, after assessing the discriminatory ability of subsets of the 32 sensors of Cyranose 320 for the six different possible combinations of the 4 studied groups; increased feasibility and modest to good diagnostic accuracy values were obtained
[114115]. Cyranose 320 has been also used for COPD phenotyping permitting (especially in the case of GOLD stage I) the detection of activation of inflammatory cells, indicating increased inflammatory activity in mild rather than severe COPD
[130131].
2.2. Cancers
Discrimination between different cancer types and cancer stages/histologies as well as between malignant and benign tumors (additionally to LC which has been mentioned earlier) using breath analysis of VOCs is also a hot research topic. Analytical techniques have been used for such applications. Phillips at al., for example, detected 5 VOCs and were able to differentiate BC patients from patients with abnormal mammograms and negative biopsies, with 93.8% sensitivity and 84.6% specificity
[137138]. Haick’s group identified 21 exhaled VOCs that were significantly different between HC and patients suffering from breast benign tumors (BBT), ductal carcinoma in situ (DCIS, early-stage BC) and BC; a potentially cancer-related set of 14 VOCs that were significantly different between malignant and non-malignant patients was also identified in this study thus permitting group differentiation with 78% sensitivity and 72% accuracy
[100101]. In another study by the same group, the detection of GCa and the presence/absence and risk level of precancerous lesions was attempted using GC-MS so as to identify 8 VOCs statistically different between GCa and operative link on gastric intestinal metaplasia (OLGIM) groups (e.g., GCa vs. OLGIM 0-IV, GCa vs. OLGIM 0-II, GCa vs. OLGIM 0), as well as between GCa and peptic ulcer disease (PUD) and OLGIM 0-IV and PUD (p-values < 0.017)
[101102]. Those 8 VOCs, in different combinations, are considered to correspond to the breathprints of OLGIM groups
[101102].
Remarkably, respective applications of sensing devices have been extensively investigated for various cancer types. Haick’s group has used NA-NOSE in order to discriminate between subjects with BC, benign breast conditions or normal mammographs, achieving increased sensitivity and specificity values for the BBT patients in comparison to the other 2 groups
[103104]. In another study by the same group a chemiresistor based on organically-coated Au NPs and SWCNTs was successfully used for the differentiation of BC from BBT and HC, BBT only or DCIS only, as well as for the differentiation of different molecular BC sub-groups as presented on
Table 3. Larger studies are necessary, though for significant statistical results and more information to be obtained
[100101]. Very recently, differentiation between BC and LC has been achieved as well by Rodriguez et al. with a correct classification of 93.05%
[4849]. Concerning GCa, the differentiation of GCa and OLGIM groups as well as between different OLGIM stages (e.g., OLGIM 0 vs. I–II, 0 vs. I–IV, I–II vs. III–IV) has been attempted by Haick’s group using the same type of chemiresistor-array and leading to high validated accuracy values in some cases (
Table 3)
[101102]. In another promising study, Haick et al. used a nanomaterial-based chemiresistor (
Table 3) that permitted the successful discrimination of Gca from benign conditions along with Gca staging (early vs. late stages)
[102103]. Gca differentiation from gastric ulcer patients has been achieved by Daniel et al. with a great classification ability, using an array of commercial MOS gas sensors and various ANN types
[105106]. Remarkably, a molecularly modified Si NW FET, developed by Haick’s group, has permitted Gca staging with an accuracy of 87% as well as Gca differentiation from LC with an accuracy of 92%
[4344]. More recently, discrimination of ovarian cancer (OC) from women with benign tumors and HC was achieved by Raspagliesi et al. with great classification performance, both in the case of strict and most probable prediction, using again MOS sensors
[106107]. Notably in class prediction application, 4/23 early-stage OC patients were misclassified as benign/HC along with 2/14 OC patients with tumor size < 3 cm in cross validation phase, while in prediction phase only 1/9 early-stage patients were misdiagnosed
[106107]. Another common cancer type, i.e., head and neck cancer (HNC), has been studied for differential diagnosis with sensing devices. As an example, Hooren et al. attempted to discriminate HNC from LC with a high accuracy of 93% analyzing patients’ exhaled breath with AeoNose, excluding cutaneous tumors and salivary glands malignancies
[9293]. HNC differentiation from colon and bladder cancer with AeoNose was also reported, by the same group along with the discrimination between bladder and colon cancer, demonstrating the discriminant ability of the e-Nose for those cancer types after double cross-validation
[9394]. More recently, Cyranose 320 was used for the differentiation of advanced bronchial (LC) and laryngeal (HNC) SCC, as well as for the discrimination of advanced and in situ stages of bronchial and laryngeal SCC, leading to successful classification of the groups (
Table 3)
[8889].
2.3. Liver, Renal, and Intestinal Diseases
Liver cirrhosis, chronic hepatitis
[108109], CKD
[4647], and inflammatory bowel diseases (IBD) comprise common liver, kidney, and intestinal diseases, respectively. As far as CKD is concerned, it is characterized by gradual loss of kidney function within months or years, while different treatment is demanded depending on disease stage (stages I–V)
[4647]. Similarly, in the case of chronic liver disease (CLD), disease staging is of great importance; following CLD diagnosis, using invasive biopsy, liver function assessment is conducted biochemically
[108109]. On the other hand, early stage and precise IBD and IBS diagnosis, as well as the invasive diagnostic methods followed for IBD, comprise challenging issues
[104105]. Consequently, precise diagnosis and staging of liver, kidney and intestinal diseases are particularly important and have been attempted using breath analysis with sensing devices. Pennazza et al. for instance used BIONOTE e-nose to successfully differentiate not only liver cirrhosis and CLD from non-cirrhotic CLD (chronic hepatitis) but also liver cirrhosis stages by taking into account smoking habits and potential comorbidities (e.g., diabetes, lung, and heart diseases)
[108109]. Concerning renal diseases, Haick’s group achieved CKD staging using organically functionalized Au NPs-based chemiresistors and SVM
[4647]. The classification of CKD stage IV in relation to stage V was permitted by 2 or 3 sensors with an accuracy of 85%, a sensitivity of 75% and specificity of 92% while only one sensor allowed for the discrimination of early and advanced stages with 76% accuracy, 75% sensitivity, and 77% specificity
[4647]. The discrimination of CKD from other diseases has been also attempted. Specifically, discrimination of CKD, diabetes, and HC with high or low creatinine has been attempted with success using an array of commercial (MQ) sensors along with different classification methods, with SVM and PCA leading to good group classification
[109110]. Remarkably, pre-concentration or dehumidification were not needed for clear classification to be accomplished
[109110]. The effective discrimination between the intestinal diseases IBD and IBS has been also reported along with further differentiation of the IBD into ulcerative colitis (UC) and Crohn’s disease (CD), using not only artificial but also real-breath samples and a MCNPs-based chemiresistor (
Table 3)
[104105]. The higher accuracy values observed when using artificial samples is expected and attributed to the standard concentration of VOCs, contrary to the variable concentration of VOCs in breath
[104105].
2.4. Neurodegenerative Diseases
Neurodegenerative diseases are characterized by gradually augmented occurrence, as a direct consequence of the increased lifespan of human population
[138139], with Alzheimer (AD) and Parkinson (PD) being the most frequent
[5051]. Concerning AD, early disease detection is of great importance for preventing, decelerating and terminating the disease
[138139] while diagnosis for both diseases is based on the assessment of clinical symptoms
[5051]. Remarkably, the analysis of exhaled VOCs has been investigated for precise AD diagnosis as well as for differential diagnosis between AD and PD
[5051][138139]. Recently, Tiele et al. attempted to discriminate mild cognitive impairment due to AD (MCI) from AD, using GC-MS, achieving 60% sensitivity and 84% specificity along with the detection of 6 potential discriminant VOCs
[138139]. Haick’s group, on the other hand, achieved the discrimination of AD and PD as well as an overall discrimination of AD, PD, and HC, using an array of 20 nanomaterial-based chemiresistors and GC-MS (
Table 3)
[5051]. In a similar study, the ability of IMS and Cyranose 320 to differentiate AD, PD, and HC was demonstrated, achieving a high overall discriminant capability (
Table 3)
[8990]. IMS analysis revealed five VOCs significantly different between the groups
[8990]. The discrimination of the same groups of subjects has been also attempted using different arrays of MOS sensors (TGS, MICS) with one of the combinations (8 MOS sensors) demonstrating the best discriminant ability
[107108].
3. Conclusions
Exhaled breath analysis, especially using selective or cross-reactive sensors, comprises a non-invasive method that holds a great promise for application in early-stage and differential diagnosis of not only respiratory but also systemic diseases. This entry presents the main categories of nanomaterials and sensors that have been used up to now in exhaled breath analysis for disease diagnosis as well as to demonstrate the applicability of breath analysis in differential diagnosis, phenotyping, and staging of several types of diseases, especially via the use of cross-reactive sensing devices.
The progressive development of novel nanomaterials offers a great opportunity to develop more effective sensing elements, both for selective and cross-reactive sensors and especially for point-of-care diagnosis, treatment monitoring and population screening. However, fundamental challenges in this novel research field inhibit the application of breath analysis in clinical practice and should therefore be addressed. Concerning analytical techniques used for exhaled VOCs identification, the use of bulky, expensive, and complex analytical devices is limited in hospitals while their incorporation in portable point-of-care systems is still unattainable
[139140]. In addition, the validity of breath analysis results is of major concern since the trace levels of exhaled VOCs affect the analysis accuracy
[140141]. At the same time the lack of clear breath sampling protocols
[140141], e.g., breath collection
[139140] and breath storage, could potentially change sample composition
[139140] and therefore emerge as important challenges. Sample composition can be also affected by confounding factors, i.e., age, gender, place of living, habits, and nutrition. In the case of sensors exhalation rate, a hardly controlled parameter, may also play a confusing role hence complicating the procedure
[139140].