Demystification of artificial intelligence for respiratory clinicians managing patients with obstructive lung diseases

Figures & data

Figure 1. Artificial intelligence hierarchy. Artificial intelligence comprises a variety of methods, including supporting technologies, rule-based expert systems and machine learning. Machine learning can be further divided into supervised, unsupervised and reinforcement learning. KNN – K-nearest neighbors; OLS – ordinary least squares; SVM – support vector machine.

Figure 2. Graphical representation of (a) classification and (b) regression problems of supervised learning; (c) clustering analysis and (d) reinforcement learning.

Table 1. Summary of characteristics of the included studies applying machine learning for asthma diagnosis and management.

First author, year	Study population	Input features	Output	Algorithms	Validation	Performance
Tomita, 2019 [23]	566 subjects (367 people with asthma, 199 healthy controls)	Age, sex, cough, history wheezing, diurnal variation of symptoms, repeated symptoms, past history of allergy diseases, family history of allergy diseases, smoking status, current wheezing on auscultation, biochemical findings, spirometry parameters, bronchial challenge test	Asthma vs healthy controls	LR, SVM, DNN	10-fold stratified cross validation	DNN: Acc = 98%
Singh, 2018 [45]	73 subjects (43 people with asthma and 30 healthy controls)	Carbon dioxide waveforms	Asthma vs healthy controls	SVM, KNN, NB	LOOCV	SVM: Acc = 95%
Islam, 2018 [20]	60 subjects (30 people with asthma and 30 healthy controls)	Respiratory sounds	Asthma vs healthy controls	ANN, SVM	LOOCV	ANN (2- channel): Acc = 89% SVM (3-channel): Acc = 93%
Ullah, 2019 [46]	202 subjects (150 people with asthma, 52 healthy controls)	Raman spectral data	Asthma vs healthy controls	ANN, SVM, RF	10-fold cross validation	SVM: Acc = 94%, AUC = 0.90
Kisiel, 2020 [47]	1,291 people with asthma	Demographic and clinical characteristics	NA	k-medoids	Reproducibility in an independent sample	3 clusters
Oh, 2020 [48]	590 people with asthma	Blood biomarkers from routine blood tests	NA	k-means	Prospective validation with acute exacerbations	3 clusters
Xiang, 2020 [50]	31,433 people with asthma	Electronic health records	Exacerbation	LR, ANN, DNN	Train-test split; 5-fold cross validation	DNN: AUC = 0.70
Zein, 2021 [49]	60,302 people with asthma	Electronic health records	Non-severe exacerbations; emergency department visits and hospitalizations	LR, RF and LightGBM	Train-Test split	Non-severe exacerbations LightGBM: AUC = 0.71 Emergency department visits LightGBM: AUC = 0.88 Hospitalizations LightGBM: AUC = 0.85
Luo, 2020 [51]	334,564 entries (315,308 from 2005–2016 and 19,256 from 2017)	235 predictors	Emergency department visits or hospitalizations	40 algorithms	Cross-validation; Dataset from 2017 used for external validation	XGBoost: AUC = 0.86
Hafke-Dys, 2021 [33]	899 people with asthma (19,166 recordings)	Indexes from a digital AI-powered stethoscope	Abnormal or normal cough	DNN	Train-Test split	AUC = 0.94
Khasha, 2019 [52]	96 people with asthma	Demographic characteristics, spirometry parameters, symptoms, medical history and environmental factors	Asthma control level (very poorly controlled, not well controlled, well controlled)	LR, SVM, RF, XGBoost, KNN, DT, NB, rule-based model from clinicians’ knowledge	5-fold cross-validation	Machine learning+rule-based model Acc = 91.7%
Halner, 2021 [53]	81 subjects (59 people with asthma; 22 people with COPD)	Demographics, medical history, medication, VAS, MRC dyspnea scale, HADS, EuroQol 5D, vital signs, lab results	Systemic Corticosteroids prescription and treatment failure	RF	LOOCV	Systemic corticosteroids Prescription AUC = 0.69 Treatment Failure AUC = 0.81

Acc: Accuracy; ANN: Artificial Neural Network; AUC: Area under the Receiver Operating Characteristic Curve; COPD: chronic obstructive pulmonary disease; DT: Decision tree; DNN: Deep Neural Network; XGBoost: extreme gradient boosting; HADS: Hospital Anxiety and Depression Scale; KNN: k-nearest neighbor; LOOCV: Leave-One-Out Cross-Validation; LightGBM: light gradient-boosting machine; LR: Logistic regression; MRC: Medical Research Council; NA: Not applicable; NB: Naïve Bayes; RF: Random Forest; SVM: Support Vector Machine; VAS: Visual Analogue Scale.

Table 2. Summary of characteristics of the included studies applying machine learning for COPD diagnosis and management.

First author, year	Study population	Input features	Output	Algorithms	Validation	Performance
Inbar, 2021 [54]	234 subjects (73 people with chronic heart failure, 75 people with COPD and 86 healthy controls)	CEPT data	Chronic heart failure vs COPD vs healthy controls	SVM	Train-test split; cross-validation	Acc = 99%
Srivastava, 2021 [24]	126 subjects	Respiratory sounds	COPD vs healthy controls	DNN	Train-test split; 10-fold cross-validation	Sn = 0.93; Sp = 0.93
Haider, 2019 [57]	55 subjects (30 people with COPD and 25 healthy controls)	Respiratory sounds and spirometry parameters	COPD vs healthy controls	SVM, KNN, LR, DT and discriminant classifiers	Train-test split; 5-fold and 10-fold cross-validation	Lung sounds SVM: Acc = 83.6% Lung sounds + spirometry SVM: Acc = 100%
Amaral, 2015 [61]	168 subjects (126 people with COPD and 42 healthy controls)	Forced oscillation technique parameters	COPD different levels of obstruction vs healthy controls	SVM (Linear and radial basis kernel), KNN, RF	10-fold cross-validation	Control versus all people with COPD and control versus moderate, severe, and very severe people with COPD separately AUC >0.90
Berkel, 2010 [58]	79 subjects (50 people with COPD, 29 healthy controls)	Volatile organic compounds identified by gas chromatography-mass spectroscopy	COPD vs healthy controls	SVM	Cross-validation, external validation (16 people with COPD and 16 controls)	Cross-validation Acc = 92% Test set Acc = 91%
V.A B, 2021 [59]	199 subjects (55 people with COPD, 51 people with lung cancer and 93 healthy controls)	Volatile organic compounds identified by an electronic nose device	COPD vs healthy controls, lung cancer vs healthy controls	XGBoost, AdaBoost, RF	Train-test split; 3-fold, 5-fold and 10-fold cross-validation	COPD vs Healthy controls XGBoost: Acc = 77%, AUC = 0.76
González, 2018 [55]	7,983 COPD smokers (8,983 from COPDGene study and 1,672 from ECLIPSE study)	Chest computed tomography imaging	COPD vs controls, 3-year all-cause mortality	DNN	Train-test split of the sample from COPDGene study, ECLIPSE sample used for external validation	COPD vs. controls AUC = 0.856 Mortality C-statistic = 0.60
Tang, 2020 [56]	4,784 COPD former and current smokers (2,589 from PanCan study and 2,195 from ECLIPSE)	Chest computed tomography imaging	COPD vs healthy controls	DNN	3-fold cross-validation, ECLIPSE sample used for external validation	AUC = 0.886
Mariani, 2021 [60]	19,077 people with asthma or COPD (2007–2011)	Electronic Health Records	COPD vs asthma vs ACOS	SVM (linear and radial basis function), KNN, RF	Train-test split; 10 to 30-fold cross-validation	pre-rec AUC Asthma = 0.9, ACOS = 0.34, COPD = 0.93, unclear = 0.25
Garcia-Aymerich, 2011 [62]	342 people with COPD hospitalized with exacerbation	Respiratory symptom and quality of life, lung function, chest x-ray morphometry, systematic inflammation, nutritional status, muscle strength, exercise capacity, comorbidities, lung density, airway morphology, bronchial colonization and inflammation data	NA	k-means	Prospective validation with hospital admission and all-cause mortality	3 clusters
Burgel, 2017 [64]	2,409 people with COPD	Age, BMI, spirometry parameters, mMRC dyspnea scale, number of previous exacerbations, presence of comorbidities	5 clusters	Hierarchical algorithm, DT	Train-test split; Prospective validation with all-cause mortality	5 clusters DT Acc = 77%
Marques, 2022 [63]	352 people with COPD	Sociodemographic and clinical data, lung function, symptoms, disease impact, quality of life, lower-limb muscle strength and functional status	4 clusters	k-medoids, DT	Train-test split; stability over time	4 clusters DT Acc = 71.7%
Vanfleteren, 2013 [26]	213 people with COPD	13 comorbidities (chronic kidney disease, anemia, hypertension, obesity, underweight, muscle wasting, hyperglycemia, dyslipidemia, osteoporosis, anxiety and depression, atherosclerosis, myocardial infarction)	NA	Self-organizing maps and hierarchical algorithm	NA	5 clusters
Augustin, 2018 [65]	518 people with COPD	Lung function attributes	NA	Self-organizing maps and hierarchical algorithm	NA	7 clusters
Cavailles, 2020 [66]	143,006 people with COPD hospitalized for an acute exacerbation	Sociodemographic and clinical variables	Rehospitalization for acute exacerbation	DT	Train-test split	NA
Fernandez-Granero, 2015 [67]	15 people with COPD	Respiratory sounds	Acute exacerbations	SVM	10-fold cross validation	Sn = 0.74, Sp = 0.98
Aguilaniu, 2021 [68]	1,409 people with COPD	Clinicopathological data	Extreme inactivity	RF	Train-test split	AUC = 0.84
Moll, 2020 [69]	3,900 people with COPD (2,632 from COPDGene study and 1,268 from ECLIPSE)	Clinical, spirometry and imaging features	All-cause mortality	Cox regression with RF	Train-test split of the sample from COPDGene study, ECLIPSE sample used for external validation	Cox regression C-statistic = 0.702 BODE C-statistic = 0.66 BODE and exacerbations C-statistic = 0.67 updated BODE C-statistic = 0.65 ADO C-statistic = 0.65
Esteban, 2011 [12]	611 people with COPD	Sociodemographic data, spirometry parameters, symptoms, physical activity and number of hospital admissions in the previous 2 years	5-year mortality	DT	External validation (348 people with COPD)	DT AUC = 0.74 BODE AUC = 0.72 HADO AUC = 0.70 ADO AUC = 0.74
Luo, 2020 [70]	263,015 people with COPD (780,295 hospital admissions)	Medical insurance data (demographic, clinical and costs information)	High-cost patients	LR, XGBoost, RF	Train-test split and external validation (2 datasets)	Train-test XGBoost: AUC = 0.80 Test set 1 XGBoost: AUC = 0.78 Test set 2 XGBoost: AUC = 0.76

Acc: Accuracy; AdaBoost: adaptive boosting; ADO: age, dyspnea and airflow obstruction; pre-rec AUC: Area under precision-recall curve; AUC: Area under the Receiver Operating Characteristic Curve; BMI: Body mass index; BODE: body mass index, airflow obstruction, dyspnea, and exercise capacity; CEPT: Cardiopulmonary Exercise Test; COPD: chronic obstructive pulmonary disease; DT: Decision tree; DNN: Deep Neural network; XGBoost: extreme gradient boosting; HADO: health, activity, dyspnea, obstruction; KNN: k-nearest neighbors; LR: Logistic regression; mMRC: modified Medical Research Council; NA: Not applicable; RF: Random Forest; Sn: Sensitivity; Sp: Specificity; SVM: Support Vector Machine.

Tomita K, Nagao R, Touge H, et al. Deep learning facilitates the diagnosis of adult asthma. Allergol Int. 2019 Oct;68(4):456–461.

 PubMed Web of Science ®Google Scholar

Singh OP, Palaniappan R, Malarvili M. Automatic quantitative analysis of human respired carbon dioxide waveform for asthma and non-asthma classification using support vector machine. IEEE Access. 2018;6:55245–55256. doi: 10.1109/ACCESS.2018.2871091

 Web of Science ®Google Scholar

Islam MA, Bandyopadhyaya I, Bhattacharyya P, et al. Multichannel lung sound analysis for asthma detection. Comput Methods Programs Biomed. 2018 Jun;159:111–123.

 PubMed Web of Science ®Google Scholar

Ullah R, Khan S, Ali H, et al. A comparative study of machine learning classifiers for risk prediction of asthma disease. Photodiagnosis Photodyn Ther. 2019 Dec;28:292–296.

 PubMed Web of Science ®Google Scholar

Kisiel MA, Zhou X, Sundh J, et al. Data-driven questionnaire-based cluster analysis of asthma in Swedish adults. NPJ Prim Care Respir Med. 2020 Apr 6;30(1):14. doi: 10.1038/s41533-020-0168-0

 PubMed Web of Science ®Google Scholar

Oh JH, Ahn KM, Chung SJ, et al. Usefulness of routine blood test-driven clusters for predicting acute exacerbation in patients with asthma. Respir Med. 2020 Aug;170:106042.

 PubMed Web of Science ®Google Scholar

Xiang Y, Ji H, Zhou Y, et al. Asthma exacerbation prediction and risk factor analysis based on a time-sensitive, attentive neural network: retrospective cohort study. J Med Internet Res. 2020 Jul 31;22(7):e16981. doi: 10.2196/16981

 PubMed Web of Science ®Google Scholar

Zein JG, Wu CP, Attaway AH, et al. Novel Machine Learning Can Predict Acute Asthma Exacerbation. Chest. 2021 May;159(5):1747–1757.

 PubMed Web of Science ®Google Scholar

Luo G, He S, Stone BL, et al. Developing a model to predict hospital encounters for asthma in asthmatic patients: secondary analysis. JMIR Med Inform. 2020 Jan 21;8(1):e16080. doi: 10.2196/16080

 PubMed Web of Science ®Google Scholar

Hafke-Dys H, Kuźnar-Kamińska B, Grzywalski T, et al. Artificial Intelligence Approach to the monitoring of respiratory sounds in asthmatic patients. Front Physiol. 2021;12:745635. doi: 10.3389/fphys.2021.745635

 PubMed Web of Science ®Google Scholar

Khasha R, Sepehri MM, Mahdaviani SA. An ensemble learning method for asthma control level detection with leveraging medical knowledge-based classifier and supervised learning. J Med Syst. 2019 Apr 26;43(6):158. doi: 10.1007/s10916-019-1259-8

 PubMed Web of Science ®Google Scholar

Halner A, Beer S, Pullinger R, et al. Predicting treatment outcomes following an exacerbation of airways disease. PLoS One. 2021;16(8):e0254425. doi: 10.1371/journal.pone.0254425

 PubMed Web of Science ®Google Scholar

Inbar O, Inbar O, Reuveny R, et al. A machine learning approach to the interpretation of cardiopulmonary exercise tests: development and validation. Pulm Med. 2021;2021:5516248–5516248. doi: 10.1155/2021/5516248

 PubMed Web of Science ®Google Scholar

Srivastava A, Jain S, Miranda R, et al. Deep learning based respiratory sound analysis for detection of chronic obstructive pulmonary disease. PeerJ Comput Sci. 2021;7:e369. doi: 10.7717/peerj-cs.369

 PubMedGoogle Scholar

Haider NS, Singh BK, Periyasamy R, et al. Respiratory sound based classification of chronic obstructive pulmonary disease: a risk stratification approach in machine learning paradigm. J Med Syst. 2019 Jun 28;43(8):255. doi: 10.1007/s10916-019-1388-0

 PubMed Web of Science ®Google Scholar

Amaral JL, Lopes AJ, Faria AC, et al. Machine learning algorithms and forced oscillation measurements to categorise the airway obstruction severity in chronic obstructive pulmonary disease. Comput Methods Programs Biomed. 2015 Feb;118(2):186–197.

 PubMed Web of Science ®Google Scholar

Van Berkel JJ, Dallinga JW, Möller GM, et al. A profile of volatile organic compounds in breath discriminates COPD patients from controls. Respir Med. 2010 Apr;104(4):557–563.

 PubMed Web of Science ®Google Scholar

B VA, Subramoniam M, Mathew L. Detection of COPD and lung cancer with electronic nose using ensemble learning methods. Clin Chim Acta. 2021 Dec 01;523:231–238.

 PubMed Web of Science ®Google Scholar

González G, Ash SY, Vegas-Sánchez-Ferrero G, et al. Disease staging and prognosis in smokers using deep learning in chest computed tomography. Am J Respir Crit Care Med. 2018 Jan 15;197(2):193–203. doi: 10.1164/rccm.201705-0860OC

 PubMed Web of Science ®Google Scholar

Tang LYW, Coxson HO, Lam S, et al. Towards large-scale case-finding: training and validation of residual networks for detection of chronic obstructive pulmonary disease using low-dose CT. Lancet Digit Health. 2020 May;2(5):e259–e267.

 PubMedGoogle Scholar

Mariani S, Metting E, Lahr MMH, et al. Developing an ML pipeline for asthma and COPD: the case of a Dutch primary care service. Int J Intell Syst. 2021;36(11):6763–6790. doi: 10.1002/int.22568

 Web of Science ®Google Scholar

Garcia-Aymerich J, Gómez FP, Benet M, et al. Identification and prospective validation of clinically relevant chronic obstructive pulmonary disease (COPD) subtypes. Thorax. 2011 May;66(5):430–7.

 PubMed Web of Science ®Google Scholar

Burgel P-R, Paillasseur J-L, Janssens W, et al. A simple algorithm for the identification of clinical COPD phenotypes. Eur Respir J. 2017;50(5):1701034. doi: 10.1183/13993003.01034-2017

 PubMed Web of Science ®Google Scholar

Marques A, Souto-Miranda S, Machado A, et al. COPD profiles and treatable traits using minimal resources: identification, decision tree and stability over time. Respir Res. 2022 Feb 14;23(1):30. doi: 10.1186/s12931-022-01954-6

 PubMed Web of Science ®Google Scholar

Vanfleteren LE, Spruit MA, Groenen M, et al. Clusters of comorbidities based on validated objective measurements and systemic inflammation in patients with chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2013 Apr 1;187(7):728–735. doi: 10.1164/rccm.201209-1665OC

 PubMed Web of Science ®Google Scholar

Augustin IML, Spruit MA, Houben-Wilke S, et al. The respiratory physiome: clustering based on a comprehensive lung function assessment in patients with COPD. PLoS One. 2018;13(9):e0201593. doi: 10.1371/journal.pone.0201593

 PubMed Web of Science ®Google Scholar

Cavailles A, Melloni B, Motola S, et al. Identification of patient profiles with high risk of hospital re-admissions for acute COPD exacerbations (AECOPD) in France using a machine learning model. Int J Chron Obstruct Pulmon Dis. 2020;15:949–962. 10.2147/COPD.S236787

 PubMed Web of Science ®Google Scholar

Fernandez-Granero MA, Sanchez-Morillo D, Leon-Jimenez A. Computerised analysis of telemonitored respiratory sounds for predicting acute exacerbations of COPD. Sensors. 2015;15(10):26978–26996. doi: 10.3390/s151026978

 PubMed Web of Science ®Google Scholar

Aguilaniu B, Hess D, Kelkel E, et al. A machine learning approach to predict extreme inactivity in COPD patients using non-activity-related clinical data. PloS One. 2021;16(8):e0255977. doi: 10.1371/journal.pone.0255977

 PubMed Web of Science ®Google Scholar

Moll M, Qiao D, Regan EA, et al. Machine learning and prediction of all-cause mortality in COPD. Chest. 2020 Sep;158(3):952–964.

 PubMed Web of Science ®Google Scholar

Esteban C, Arostegui I, Moraza J, et al. Development of a decision tree to assess the severity and prognosis of stable COPD. Eur Respir J. 2011;38(6):1294–300. doi: 10.1183/09031936.00189010

 PubMed Web of Science ®Google Scholar

Luo L, Li J, Lian S, et al. Using machine learning approaches to predict high-cost chronic obstructive pulmonary disease patients in China. Health Inform J. 2020 Sep;26(3):1577–1598.

 PubMed Web of Science ®Google Scholar