Conflict of interests

20
Analysis of Exhaled Breath with Analysis of Exhaled Breath with Electronic Nose and Diagnosis of Lung Electronic Nose and Diagnosis of Lung Cancer by Support Vector Machine Cancer by Support Vector Machine Dr.med. Dr.med. Māris Bukovskis Māris Bukovskis 1 2 3 1 2 3 , Dr.biol. Gunta , Dr.biol. Gunta Strazda Strazda 1 2 3 1 2 3 , Dr.med Uldis Kopeika , Dr.med Uldis Kopeika 3 4 3 4 , , Dr.biol.Normunds Jurka Dr.biol.Normunds Jurka 3 , Dr. Ainis Pirtnieks , Dr. Ainis Pirtnieks 4 , , Ph.dr. Līga Balode Ph.dr. Līga Balode 3 , Dr. Jevgenija Aprinceva , Dr. Jevgenija Aprinceva 2 , , Inara Kantane Inara Kantane 5 , Prof. Immanuels Taivans , Prof. Immanuels Taivans 1 2 3 1 2 3 1 1 Center of Lung Diseases, Pauls Stradins Clinical University Hospital, Center of Lung Diseases, Pauls Stradins Clinical University Hospital, 2 2 Faculty of Medicine, University of Latvia, Faculty of Medicine, University of Latvia, 3 3 Institute of Experimental and Clinical Medicine, University of Latvia, Institute of Experimental and Clinical Medicine, University of Latvia, 4 Department of Thoracic Surgery, Pauls Stradins Clinical University Hospital Department of Thoracic Surgery, Pauls Stradins Clinical University Hospital 5 Faculty of Economics and Management, University of Latvia Faculty of Economics and Management, University of Latvia

description

Analysis of Exhaled B reath with Electronic N ose and Diagnosis of Lung C ancer by Support V ector M achine. - PowerPoint PPT Presentation

Transcript of Conflict of interests

Page 1: Conflict of interests

Analysis of Exhaled Breath with Electronic Nose and Analysis of Exhaled Breath with Electronic Nose and Diagnosis of Lung Cancer by Support Vector Diagnosis of Lung Cancer by Support Vector

MachineMachineDr.med. Dr.med. Māris BukovskisMāris Bukovskis 1 2 31 2 3, Dr.biol. Gunta Strazda , Dr.biol. Gunta Strazda 1 2 31 2 3, Dr.med , Dr.med Uldis Kopeika Uldis Kopeika 3 43 4, Dr.biol.Normunds Jurka , Dr.biol.Normunds Jurka 33, Dr. Ainis Pirtnieks , Dr. Ainis Pirtnieks 44, , Ph.dr. Līga Balode Ph.dr. Līga Balode 33, Dr. Jevgenija Aprinceva , Dr. Jevgenija Aprinceva 22, Inara Kantane , Inara Kantane 55, ,

Prof. Immanuels Taivans Prof. Immanuels Taivans 1 2 31 2 3

1 1 Center of Lung Diseases, Pauls Stradins Clinical University Hospital, Center of Lung Diseases, Pauls Stradins Clinical University Hospital, 2 2 Faculty of Medicine, University of Latvia, Faculty of Medicine, University of Latvia, 3 3 Institute of Experimental and Clinical Medicine, University of Latvia, Institute of Experimental and Clinical Medicine, University of Latvia, 44 Department of Thoracic Surgery, Pauls Stradins Clinical University Hospital Department of Thoracic Surgery, Pauls Stradins Clinical University Hospital55 Faculty of Economics and Management, University of Latvia Faculty of Economics and Management, University of Latvia

Page 2: Conflict of interests

Conflict of interestsConflict of interests

• No conflict of interestsNo conflict of interests

• Study was sponsored by ERAF activity 2.1.1.1.0 Study was sponsored by ERAF activity 2.1.1.1.0 Project Nb. 2010/0303/2DP/2.1.1.1.0/10/APIA/Project Nb. 2010/0303/2DP/2.1.1.1.0/10/APIA/ VIAA/043/VIAA/043/

Page 3: Conflict of interests

Lung cancer mortality and diagnostic methodsLung cancer mortality and diagnostic methods

• Lung cancer causesLung cancer causes 1.3 million deaths annually, 1.3 million deaths annually, more more than the next three most common cancers (colon, breast than the next three most common cancers (colon, breast and prostate)and prostate) combinedcombined

• 58 - 73% of patients with stage I lung cancer survive for 58 - 73% of patients with stage I lung cancer survive for 5 years5 years

• For distant tumors the For distant tumors the 55-year survival rate is only 3.5 -year survival rate is only 3.5 %%

• Available diagnostic methods - nonsensitive, expensive Available diagnostic methods - nonsensitive, expensive or invasiveor invasive

World Health Organization. CancerWorld Health Organization. Cancer Fact Sheet Fact Sheet 2009 2009American Cancer Society.American Cancer Society. Cancer Facts & Figures 201 Cancer Facts & Figures 2012

Page 4: Conflict of interests

VOC’s in exhaled breathVOC’s in exhaled breath

Lung cancer sniffer dogsLung cancer sniffer dogsCBC News Aug 17, 2011CBC News Aug 17, 2011

Gordon SM et al. Clin Chem 1985Gordon SM et al. Clin Chem 1985Machado et al. AJRCCM 2005Machado et al. AJRCCM 2005

Chen X et al. Cancer 2007Chen X et al. Cancer 2007

Page 5: Conflict of interests

e- e- e-

e- e-e-

Functional principles of electronic noseFunctional principles of electronic nose

S1 S2 S3 S4 S5 S6

• VOCs induce change of the sensorVOCs induce change of the sensorvolume and subsequently change of volume and subsequently change of electric resistanceelectric resistance

• A unique response curve combination, A unique response curve combination, containing the information to allow containing the information to allow discrimination of the different samplesdiscrimination of the different samples

Cyranose 320Cyranose 320

VOCsVOCs

Page 6: Conflict of interests

ObjectiveObjective

• The aim of our study was to prove the potential of The aim of our study was to prove the potential of exhaled breath analysisexhaled breath analysis and Support Vector Machine and Support Vector Machine (SVM)(SVM) to discriminate to discriminate patients with: patients with:

1) 1) lung cancer from healthy controls and other lung lung cancer from healthy controls and other lung diseasesdiseases;;

2) 2) lung cancer with or without COPD from patients with lung cancer with or without COPD from patients with only COPD and healthy controlsonly COPD and healthy controls;;

3) early 3) early stage lung cancerstage lung cancer..

Page 7: Conflict of interests

MethodsMethods

Sampling of exhaled airSampling of exhaled air

• Inspiration of VOC-filtered air by tidal breathing for 5 Inspiration of VOC-filtered air by tidal breathing for 5 minutes, through T-shaped two-way non-rebreathing minutes, through T-shaped two-way non-rebreathing valve (valve (Hans Rudolph Inc., Shawnee, USAHans Rudolph Inc., Shawnee, USA))

• Inhalation to total lung capacity and full exhalation with Inhalation to total lung capacity and full exhalation with approximate flow rate 0.25 – 0.5 L/s into a polyethylene approximate flow rate 0.25 – 0.5 L/s into a polyethylene terephthalate plastic bag terephthalate plastic bag

• Analysis by electronic nose device (Analysis by electronic nose device (Cyranose 320, Smith Cyranose 320, Smith Detection, USADetection, USA) within 5 minutes after breath sample ) within 5 minutes after breath sample collectioncollection

Dragonieri S et al. J Allergy Clin Immunol 2007Dragonieri S et al. J Allergy Clin Immunol 2007

Page 8: Conflict of interests

MethodsMethods

Satistical analysisSatistical analysis

Support vector machine (SVM)Support vector machine (SVM)

• Continuous predictors: relative maximum (RContinuous predictors: relative maximum (Rmaxmax), area ), area

under curve (∑under curve (∑0-60”0-60”) and tg ) and tg αα0-60”0-60” for each curve of 32 for each curve of 32

sensorssensors

• Additional predictor factors: Additional predictor factors: age, age, smoking status smoking status (smoker, non-smoker, ex-smoker), (smoker, non-smoker, ex-smoker), smoking historysmoking history ((pack-yearspack-years)) and and ambient ambient temperature temperature ttº Cº C at the at the moment of measurementmoment of measurement

Page 9: Conflict of interests

Support Vector MachineSupport Vector Machine

Page 10: Conflict of interests

ResultsResults

Morphologically confirmed lung cancerMorphologically confirmed lung cancerOther diseases: COPD, pneumonia, tbc, PATE, benign tumors etc.Other diseases: COPD, pneumonia, tbc, PATE, benign tumors etc.Control – healthy volunteers, postinflammatory pneumofibrosisControl – healthy volunteers, postinflammatory pneumofibrosis

Page 11: Conflict of interests

ResultsResults

Classification summary Classification summary (Support Vector Machine), (Support Vector Machine), Cancer vs No cancerCancer vs No cancer, ,

Training/Test sample 100% Training/Test sample 100% SVM: SVM: Classification type Classification type 1 (C=2.000), 1 (C=2.000), KernelKernel: : Linear Linear

Number of support vectors Number of support vectors = 219 (170 = 219 (170 boundedbounded) ) Include criteriaInclude criteria: v20='GF': v20='GF'

TotalTotal CorrectCorrect IncorrectIncorrect Correct Correct (%)(%) Incorrect Incorrect (%)(%)

CancerCancer 165165 144144 2121 87.387.3 12.712.7

No cancerNo cancer 170170 121121 4949 71.271.2 28.828.8

Cancer vs No cancerCancer vs No cancer

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and ambient tºC Age, Pack-years and ambient tºC

Cross-validation 72.8% Class accuracy 79.1%Cross-validation 72.8% Class accuracy 79.1%

   CancerCancer No cancerNo cancer   

CancerCancer 144144 4949 74.674.6 PPVPPV

No cancerNo cancer 2121 121121 85.285.2 NPVNPV

87.387.3 71.271.2

SensitivitySensitivity SpecificitySpecificity

Page 12: Conflict of interests

ResultsResults

Classification summary Classification summary (Support Vector Machine), (Support Vector Machine), Cancer vs No cancerCancer vs No cancer,, Training Training

sample 75% Test sample 25%sample 75% Test sample 25% SVM: SVM: Classification type Classification type 1 (C=2.000), 1 (C=2.000), KernelKernel: : LinearLinear

Number of support vectors Number of support vectors = 219 (170 = 219 (170 boundedbounded) ) Include criteriaInclude criteria: v20='GF': v20='GF'

TotalTotal CorrectCorrect IncorrectIncorrect Correct Correct (%)(%) Incorrect Incorrect (%)(%)

CancerCancer 4545 4040 55 88.988.9 11.111.1

No cancerNo cancer 3939 2626 1313 66.766.7 33.333.3

Cancer vs No cancerCancer vs No cancer

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and ambient tºC Age, Pack-years and ambient tºC

Cross-validation 69.7% Class accuracy 75.5%Cross-validation 69.7% Class accuracy 75.5%

   CancerCancer No cancerNo cancer   

CancerCancer 4040 1313 75.575.5 PPVPPV

No cancerNo cancer 55 2626 83.983.9 NPVNPV

88.988.9 66.766.7

SensitivitySensitivity SpecificitySpecificity

Page 13: Conflict of interests

ResultsResults

Classification summary Classification summary (Support Vector Machine) (Support Vector Machine) Cancer vs Control Training/Test Cancer vs Control Training/Test

sample 100% sample 100% SVM: SVM: Classification type Classification type 1 (C=5.000), Kernel: 1 (C=5.000), Kernel: LinearLinear Number of support Number of support

vectors vectors = 84 (39 = 84 (39 boundedbounded) ) Include criteriaInclude criteria: v20='GF': v20='GF'

TotalTotal CorrectCorrect IncorrectIncorrect Correct Correct (%)(%) Incerrect Incerrect (%)(%)

CancerCancer 166166 164164 22 98.898.8 1.21.2

ControlControl 7979 6464 1515 81.081.0 19.019.0

Cancer vs ControlCancer vs Control

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and ambient tºC Age, Pack-years and ambient tºC

Cross-validation 90.6% Class accuracy 93.1%Cross-validation 90.6% Class accuracy 93.1%

   CancerCancer ControlControl   

CancerCancer 164164 1515 91.691.6 PPVPPV

ControlControl 22 6464 97.097.0 NPVNPV

98.898.8 81.081.0

SensitivitySensitivity SpecificitySpecificity

Page 14: Conflict of interests

ResultsResults

Classification summary Classification summary (Support Vector Machine) (Support Vector Machine) Cancer vs Control Training sample Cancer vs Control Training sample

75% Test sample 25% 75% Test sample 25% SVM: SVM: Classification type Classification type 1 (C=5.000), Kernel: 1 (C=5.000), Kernel: LinearLinear Number of Number of

support vectors support vectors = 84 (39 = 84 (39 saistītisaistīti) ) Include criteriaInclude criteria: v20='GF': v20='GF'

TotalTotal CorrectCorrect IncorrectIncorrect Correct Correct (%)(%) Incorrect Incorrect (%)(%)

CancerCancer 4545 4444 11 97.897.8 2.22.2

ControlControl 1616 1111 55 68.868.8 31.231.2

Cancer vs ControlCancer vs Control

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and ambient tºC Age, Pack-years and ambient tºC

Cross-validation 89.7% Class accuracy 93.5%Cross-validation 89.7% Class accuracy 93.5%

   CancerCancer ControlControl   

CancerCancer 4444 55 89.889.8 PPVPPV

ControlControl 11 1111 91.791.7 NPVNPV

97.897.8 68.868.8

SensitivitySensitivity SpecificitySpecificity

Page 15: Conflict of interests

ResultsResults

Cancer vs Cancer + COPD vs COPD vs ControlCancer vs Cancer + COPD vs COPD vs Control

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and Ambient tºC Age, Pack-years and Ambient tºC

Cross-validation 71.1% Class accuracy 77.4%Cross-validation 71.1% Class accuracy 77.4%

Classification summary Classification summary (Support Vector Machine), (Support Vector Machine), Cancer vs Cancer+COPD vs COPD vs Cancer vs Cancer+COPD vs COPD vs

ControlControl, , SVM: SVM: Classification type Classification type 1 (C=2.000), Kernel: 1 (C=2.000), Kernel: LinearLinear Number of support vectors Number of support vectors = =

152 (43 152 (43 bounded)bounded) Include criteriaInclude criteria: v20='GF': v20='GF'

TotalTotal CorrectCorrect IncorrectIncorrect Correct Correct (%)(%) Incorrect Incorrect (%)(%)

CancerCancer 6363 3636 2727 57.157.1 42.942.9

Cancer + COPDCancer + COPD 7799 7979 00 100.0100.0 0.00.0

COPDCOPD 1515 55 1010 33.333.3 66.766.7

ControlControl 7878 6262 1616 79.579.5 20.520.5

Page 16: Conflict of interests

ResultsResults

Cancer vs Cancer + COPD vs COPD vs ControlCancer vs Cancer + COPD vs COPD vs Control

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and Ambient tºC Age, Pack-years and Ambient tºC

Cross-validation 71.1% Class accuracy 77.4%Cross-validation 71.1% Class accuracy 77.4%

Classification matrix Classification matrix (Support Vector Machine), (Support Vector Machine), Cancer vs Cancer+COPD vs COPD vs Cancer vs Cancer+COPD vs COPD vs

ControlControl, , Training/Test sample 100% Training/Test sample 100% SVM: SVM: Classification typeClassification type1 (C=2.000), Kernel: 1 (C=2.000), Kernel: LinearLinear, ,

Number of support vectors Number of support vectors = 152 (43 = 152 (43 bounded) Prognosis bounded) Prognosis (r(rowsows) x ) x DiagnosisDiagnosis ( (columnscolumns))

CancerCancer Cancer + COPDCancer + COPD COPDCOPD ControlControl

CancerCancer 3636 2626 00 11

Cancer + COPDCancer + COPD 00 79 (!)79 (!) 00 00

COPDCOPD 11 99 55 00

ControlControl 77 99 00 6262

Page 17: Conflict of interests

ResultsResults

Parameters of 32 detectors RParameters of 32 detectors Rmaxmax, ∑, ∑0-600-60 un tg un tg αα0-600-60 Age, Pack-years and Ambient tºC Age, Pack-years and Ambient tºC

Patients with post-obstructive pneumonia in cancer group and bacterial, Patients with post-obstructive pneumonia in cancer group and bacterial, TB and infarct pneumonia in no cancer group were excluded from analysisTB and infarct pneumonia in no cancer group were excluded from analysis

Classification matrix Classification matrix (Support Vector Machine), (Support Vector Machine), Stage 1-2, 3 and 4Stage 1-2, 3 and 4 Training/Test Training/Test

group 100% group 100% SVM: SVM: Classification type Classification type 1 (C=1.000), Kernel: 1 (C=1.000), Kernel: LinearLinear, , Number of support Number of support

vectors vectors = 184 (73 = 184 (73 boundedbounded) ) PrognosisPrognosis (r (rowsows) x ) x DiagnosisDiagnosis ( (columns)columns)

No cancerNo cancer Stage 1-2Stage 1-2 Stage 3Stage 3 Stage 4Stage 4

No cancerNo cancer 100100 00 77 22

Stage 1-2Stage 1-2 1111 33 2525 11

Stage 3Stage 3 99 00 4040 00

Stage 4Stage 4 99 77 2727 33

Page 18: Conflict of interests

ConclusionsConclusions

Exhaled breath analysis by electronic nose using support Exhaled breath analysis by electronic nose using support vector pattern recognition method vector pattern recognition method is able to discriminate:is able to discriminate:

• LLung cancer from healthy subjects and patients with ung cancer from healthy subjects and patients with different lung diseasesdifferent lung diseases

• An An early stage lung cancer from healthy subjects and early stage lung cancer from healthy subjects and patients with different lung diseasespatients with different lung diseases

• Some Some discrimination pattern between lung cancer,discrimination pattern between lung cancer, patients with lung cancer and COPD,patients with lung cancer and COPD, COPD and control, COPD and control, even in patients with combined diseaseeven in patients with combined diseasess

Page 19: Conflict of interests

AcknowledgementsAcknowledgements

• To my colleagues and our teamTo my colleagues and our team

Prof. Immanuels TaivansProf. Immanuels Taivans

Dr.biol. Gunta StrazdaDr.biol. Gunta Strazda

Dr. Ainis PirtnieksDr. Ainis Pirtnieks

Dr.med. Uldis KopeikaDr.med. Uldis Kopeika

Dr.biol. Normunds JurkaDr.biol. Normunds Jurka

Ph.dr. Liga BalodePh.dr. Liga Balode

Doctoral student Agnese KislinaDoctoral student Agnese Kislina

Mrs. Inara KantaneMrs. Inara Kantane

Page 20: Conflict of interests

Thank You for Your Attention!Thank You for Your Attention!

How to sniff out the disease?How to sniff out the disease?