Abstract
Background
The diagnosis of interstitial lung disease (ILD) can pose a challenge as the pulmonary function test (PFT) is only minimally affected at the onset. To improve early diagnosis, this study aims to explore the potential of artificial intelligence (AI) software in assisting pulmonologists with PFT interpretation for ILD diagnosis. The software provides an automated description of PFT and disease probabilities computed from an AI model.
Study methods
In study phase 1, a cohort of 60 patients, 30 of whom had ILD, were retrospectively diagnosed by 25 pulmonologists (8 junior physicians and 17 experienced pneumologists) by evaluating a PFT (body plethysmography and diffusion capacity) and a short medical history. The experts screened the cohort twice, without and with the aid of AI (ArtiQ.PFT, V.1.4.0, ArtiQ, BE) software and provided a primary diagnosis and up to three differential diagnoses for each case. In study phase 2, 19 pulmonologists repeated the protocol after using ArtiQ.PFT for 4–6 months.
Results
Overall, AI increased the diagnostic accuracy for various lung diseases from 41.8% to 62.3% in study phase 1. Focusing on ILD, AI improved the detection of lung fibrosis as the primary diagnosis from 42.8% without AI to 72.1% with AI (p<0.0001). Phase 2 yielded a similar outcome: using AI increased ILD diagnosis based on primary diagnosis (53.2% to 75.1%; p<0.0001). ILD detections without AI support significantly increased between phase 1 and phase 2 (p=0.028) but not with AI (p=0.24).
Interpretation
This study shows that AI-based decision support on PFT interpretation improves accurate and early ILD diagnosis.
Keywords: Idiopathic pulmonary fibrosis, Rare lung diseases
WHAT IS ALREADY KNOWN ON THIS TOPIC
Patients with interstitial lung diseases often have considerable diagnostic delays. However, a prompt diagnosis could potentially improve outcomes through early initiation of appropriate management. The aim of this study is, therefore, to evaluate the added value of artificial intelligence (AI) software on the diagnostic accuracy for diagnosing interstitial lung diseases.
WHAT THIS STUDY ADDS
This is the first study that demonstrates that AI-supported interpretation of pulmonary function tests improves the detection of lung fibrosis.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
AI-guided interpretation of pulmonary function tests significantly increases the identification of interstitial lung disease and thus provides physicians an important opportunity to initiate further diagnostic approaches and to treat patients without diagnostic delay that might improve the outcome of patients with interstitial lung disease.
Introduction
Artificial intelligence (AI) is a rapidly evolving technology that can aid physicians in the diagnosis and therapeutic decisions of various diseases. Particularly in respiratory diseases, AI gains importance by supporting the interpretation of CT scans or lung function tests which present essential diagnostic tools in lung diseases.1 Computer-aided tools that facilitate pulmonary function test (PFT) interpretations have been available for decades.2 3
Those PFTs are often the first step in diagnosing most respiratory diseases when patients complain about respiratory symptoms like dyspnoea or persistent cough. Therefore, a precise interpretation is important to make correct diagnoses or to initiate the appropriate further diagnostic steps. However, visual interpretation of PFTs is influenced by inter-rater variability due to missing uniformity of the guidelines or lack of training. Particularly for physicians who do not interpret lung function every day, assessing lung function may become challenging, in particular with the identification of subtle patterns.
Automated analysis of PFT follows a rigid algorithm based on population estimates and can only distinguish between normal lung function or obstructive and restrictive ventilatory disorders. In recent years, an AI model was introduced for PFT interpretation that provides a complex diagnosis. ArtiQ.PFT provides an automated description of lung function compliant with the latest European Respiratory Society/American Thoracic Society (ERS/ATS) standards,4 disease probabilities for eight respiratory conditions that are calculated based on a machine learning model, and further suggestions that can be considered to confirm or deepen the diagnosis. Such an AI tool was reported to increase the diagnostic accuracy of physicians.5,7 In a validation study with 50 patient cases and 120 pulmonologists, ArtiQ.PFT has shown a diagnostic accuracy of 82% in correctly identifying the most probable disease compared with 46% by individual pulmonologists.6 A recent study including 24 patient cases and 2 groups of pulmonologists has further shown that pulmonologists perform better on PFT interpretation when supported by the AI software.7
Particularly in subjects with restrictive ventilatory disorders, AI-based PFT interpretation may be very helpful, as it is known that diseases accompanied by restrictive patterns are more difficult to diagnose in comparison to healthy subjects and patients with chronic obstructive pulmonary disease (COPD).3 Patients with fibrotic interstitial lung disease (ILD) often have considerable diagnostic delay, but a prompt diagnosis could potentially improve outcomes through early initiation of appropriate management.8 By using an AI-based PFT interpretation, for example, patients with ILD could be diagnosed without delay. Getting a prompt and accurate diagnosis is crucial for proper treatment of ILDs. The aim of this study is, therefore, to evaluate the added value of AI software on the diagnostic accuracy for diagnosing ILD specifically.
Methods
Study design
In this multicentre study, 60 cases of healthy or ill subjects were independently evaluated by 25 physicians based on PFT (spirometry, body plethysmography, diffusion capacity) and a short medical history (including respiratory symptoms and smoking status) (online supplemental figure S1).
The study participants were asked to review the cohort twice: once without AI support and the second time with the assistance of AI (ArtiQ.PFT) (online supplemental figure S2). They had to choose between healthy or one of the following diagnoses: ILD, COPD, asthma, neuromuscular disease, thoracic deformity (eg, kyphoscoliosis, pneumonectomy), pulmonary vascular disease or ‘other obstructive disease’ (eg, bronchiectasis, cystic fibrosis). In addition to a main diagnosis, one to three differential diagnoses could be made. Confidence in their decision was given by a Likert scale: from 1 point (‘absolutely not sure’) to 5 points (‘absolutely sure’).
After this first retrospective analysis (phase 1), the study centres received the ArtiQ.PFT software to be used in their daily clinic. After 6 months of experience with ArtiQ.PFT software, the 25 participants were invited to review the 60 patient cases again—with and without the ArtiQ.PFT software (phase 2). This second phase was added to the study to investigate if there was a learning effect after using the software in the daily clinic and if this would affect diagnostic accuracy and/or decisions.
Study participants
25 physicians from 10 pulmonology departments in non-university hospitals (4 in Austria, 5 in Switzerland and 1 in Romania) were volunteers. They had a different level of education, with 8 participants being in training and 17 participants already being pulmonologists (table 1). Additionally, only one physician had prior experience with AI. The physicians were blinded to the diagnosis of the patients from whom PFTs were analysed, and they were not informed that the study was specifically centred around early detection of ILD. 19 out of those 25 participants agreed to be involved in phase 2.
Table 1. Characteristics of physicians participating in the study.
| Physicians | n |
|---|---|
| Years of experience | |
| Training (no specialty) | 8 |
| 1–5 years as a pneumologist | 3 |
| 5–10 years as a pneumologist | 7 |
| >10 years as pneumologist | 7 |
| Experience with AI prior to study entry | |
| No | 24 |
| Yes | 1 |
AI, artificial intelligence.
Pulmonary function tests
PFTs were collected from 60 subjects (all above 50 years) at the outpatient clinic of University Hospital Leuven (Belgium) between August 2017 and May 2019. All participants were Caucasian adults aged over 18 years, each with a clinically established diagnosis. The diagnosis was based on a thorough assessment, including medical history, clinical examination, PFT, laboratory investigations, imaging and histological analysis when deemed necessary. The definitive diagnosis was subsequently confirmed by an expert panel in Leuven in accordance with current clinical guidelines.8,12 This cohort was enriched for ILD patients representing 50% of the sample; for these patients the diagnosis was confirmed by multidisciplinary discussion.
During the present study, limited clinical information (including symptoms and smoking status) and a complete PFT were provided to the pulmonologists. Out of the 60 subjects, 30 were diagnosed with ILD, 7 with COPD, 4 with neuromuscular disease, 3 with asthma, 3 with thoracic deformity (eg, kyphoscoliosis), 2 with pulmonary vascular disease and 1 with other obstructive disease. 10 subjects were healthy and had a normal PFT. The patients with ILD presented with a mean forced vital capacity of 83%±20%, a mean total lung capacity (TLC) of 76%±17% and a mean diffusion capacity (carbon monoxide diffusion capacity, DLCO) of 47%±10%. Seven were never smokers, one active smoker with a pack year value of 4 and 22 were former smokers with a pack year value of 15.6±13.0. Clinical characteristics of the 60 subjects whose PFT was analysed are presented in table 2.
Table 2. Characteristics of the 60 patients whose PFTs were retrospectively analysed.
| ILD (n=30) |
COPD (n=7) |
Asthma (n=3) |
Other obstructive disease (n=1) |
Thoracic deformity (n=3) |
Neuro muscular disease (n=4) | Pulmonary vascular disease (n=2) |
Healthy (n=10) |
|
|---|---|---|---|---|---|---|---|---|
| Age (years, mean±SD) | 69.9±7.4 | 72.9±9.6 | 64.0±7.8 | 68 | 68.01±1.0 | 61.5±3.8 | 53.0±4.2 | 64.2±5.9 |
| Sex (male:female) | 10:20 | 1:6 | 1:2 | 0:1 | 1:2 | 1:3 | 1:1 | 6:4 |
| FEV1/FVC (%, mean±SD) |
79.4±7.8 | 45.6±14.8 | 55.1±12.0 | 61.2 | 74.6±9.8 | 75.2±7.6 | 76.2±8.9 | 79.3±3.4 |
| FVC (%, mean±SD) |
83.1±20.1 | 90.5±17.8 | 104.8±8.4 | 78.5 | 90.8±25.6 | 63.6±31.3 | 83.0±1.2 | 101.6±8.0 |
| FVC (%, range) |
51–125 | 68–124 | 96–112 | – | 75–120 | 30–100 | 82–84 | 89–112 |
| FEV1 (%, mean±SD) |
85.2±17.9 | 55.3±25.8 | 75.5±21.9 | 63.7 | 86.3±11.0 | 61.4±31.4 | 79.2±7.7 | 103.6±8.3 |
| TLC (%, mean±SD) |
75.8±17.1 | 132.0±33.2 | 113.3±4 | 93.0 | 88.3±23.6 | 82.8±19.8 | 95±9.9 | 103.4±7.7 |
| DLCO (%, mean±SD) |
47.1±10.1 | 55.4±27.2 | 109.7±17.8 | 76.0 | 71.0±9.9 | 83.3±9.0 | 67.0±2.8 | 88.1±18.2 |
| DLCO (%, range) |
26–68 | 31–90 | 97–130 | – | 60–79 | 72–93 | 65–69 | 70–129 |
Per cent predicted values according to GLI 2012 spirometry reference equations.19
COPD, chronic obstructive pulmonary disease; DLCO, carbon monoxide diffusion capacity; FEV1, forced expiratory volume in 1 s; FVC, forced vital capacity; GLI, Global Lung Function Initiative; ILD, interstitial lung disease; PFT, pulmonary function test; TLC, total lung capacity.
AI software
To support the expert diagnosis, AI-based software has been used (ArtiQ.PFT V.1.8.0, ArtiQ NV, Leuven, Belgium) that automatically provides a report to support PFT interpretation and clinical decision-making.
Statistical analysis
To compare the accuracy of diagnosis without and with the support of AI in phase 1 and phase 2, a McNemar test has been used. The purpose of this test is to determine whether there is a significant difference between paired categorical data collected under two different conditions (with AI vs no AI) at different time points (phase 1 vs phase 2). A p value of 5% was used as the threshold for determining statistical significance.
Additionally, we compared the performance between phase 1 and phase 2 to explore the presence of any learning effect. Statistical analysis was conducted using a paired t-test with a significance threshold set at 5%.
Moreover, we compared the performance between less experienced junior physicians (in training, no specification) and pneumologists to investigate whether AI supports PFT interpretation independently of the level of education and experience.
Results
The diagnostic yield of the AI in this study of 60 cases is 81% based on the highest probability. When we focus on the ILD cases (30 cases from those 60), the diagnostic accuracy of the AI versus the true diagnosis is 90%.
Diagnosis without and with AI
In the first phase of the study, 25 clinicians interpreted 60 subject cases resulting in 1500 evaluations without AI and 1500 with AI support (table 3). Overall, without AI, there was an agreement between the true diagnosis of the patients and the main diagnosis made by the study participants based on complete PFT and medical history in 41.8%. With the use of ArtiQ.PFT, the agreement increased to 62.3% (relative improvement from baseline of 49.0%). When looking only at subjects suffering from ILD (table 4), without AI, the diagnostic agreement was 42.8% vs 72.1% with AI (relative improvement of 68.5%) resulting in a statistically significant difference (p<0.0001). The results are presented in figure 1.
Table 3. Patient case evaluation without and with AI prior to software implementation (phase 1) and 6 months following software implementation (phase 2).
| PFT evaluation without AI | PFT evaluation with AI | |
|---|---|---|
| Patient cases evaluation (phase 1) | ||
| Agreement with true diagnosis | 41.8% (628/1500) |
62.3% (935/1500) |
| Patient case evaluation after 6 months of AI implementation (phase 2) | ||
| Agreement with true diagnosis | 46.5% (530/1140) |
65.6% (748/1140) |
AI, artificial intelligence; PFT, pulmonary function test.
Table 4. ILD diagnosis by patient case evaluation without and with AI prior to software implementation (phase 1) and 6 months following software implementation (phase 2).
| PFT evaluation without AI | PFT evaluation with AI | |
|---|---|---|
| Patient cases evaluation (phase 1) | ||
| Agreement with true diagnosis | 42.8% (321/750) |
72.1% (541/750) |
| Patient case evaluation after 6 months of AI implementation (phase 2) | ||
| Agreement with true diagnosis | 53.2% (303/570) |
75.1% (428/570) |
AI, artificial intelligence; ILD, interstitial lung disease; PFT, pulmonary function test.
Figure 1. Diagnostic accuracy without and with AI (phase studies 1 and 2). AI, artificial intelligence; ILD, interstitial lung disease.
Phase 2 began after 6 months of experience with the AI software. There were 19 out of the 25 clinicians who made a total of 1140 PFT interpretations with and without AI. The support of AI significantly improved the accuracy from 46.5% to 65.6% overall (relative improvement from baseline of 41.1%) and from 53.2% to 75.1% for ILD subjects (relative improvement of 41.2%) (both p<0.001).
The accuracy per user had also been investigated. Overall, a statistically significant improvement was observed when using AI support (p<0.0001). Indeed, only one physician did not benefit from AI support. Differences in diagnostic accuracy between phase 1 and phase 2 were further investigated to identify a potential learning effect. The paired t-test showed a significant difference between phase 1 and phase 2 without AI for ILD diagnosis (42.8% vs 53.2%, p=0.028) but this difference was not significant between phase 1 and phase 2 with AI (72.1% vs 75.1%, p=0.24).
Differential diagnosis
In phase 1, a second and third differential diagnoses were made by, respectively, 1001 and 391 study participants without AI and by, respectively, 1000 and 371 participants with AI. Fewer differential diagnoses were given in phase 2: without AI, second and third differential diagnoses were made by 759 and 275 physicians, with AI by 776 and 291 study participants, which was proportional to the different number of physicians participating (25 vs 19).
Confidence Likert scores
In study phase 1, participants were more confident with their diagnosis when using AI software compared with PFT interpretation without AI (Likert scores 4 and 5 increased from 40.7% to 48.8%). This effect was not seen in study phase 2.
Comparison between junior physicians and experienced pneumologists
Junior physicians had significantly lower diagnostic accuracy without AI compared with specialists. When using AI, the increase in accuracy for ILD diagnosis was more pronounced in junior doctors compared with experienced pneumologists when using AI (figure 2).
Figure 2. Comparison between junior physicians (in training, no specialisation) and experienced pneumologists. AI, artificial intelligence; PFT, pulmonary function test.
Discussion
Recognition of ILD and immediate therapeutic intervention is important as it possibly slows down or even stops the progression. Diagnostic delay is a common problem for patients with ILD. A case-control study by Hewson et al reported that some patients with idiopathic pulmonary fibrosis (IPF) may be symptomatic for more than 5 years before diagnosis. These IPF patients were significantly more likely to have a diagnosis of heart failure and a diagnosis of COPD than an ILD, whereby it is unclear whether heart failure or COPD reflects coexistent diagnoses or misdiagnoses of common conditions with similar symptoms. This result demonstrates that ILD is often an undiagnosed or late-diagnosed disease.13 Non-specific symptoms that overlap with those of other more common diseases often lead to misdiagnosis or a delayed referral to a specialist centre.8
Besides medical history and laboratory tests, lung function tests including body plethysmography and the measurement of diffusion capacity, and high-resolution CT (HRCT) are the most important steps in diagnosing ILD. Particularly, the HRCT plays an important role in the investigation of patients with ILD. However, a CT scan as a screening examination for ILD has so far only been recommended for patients with systemic sclerosis.14 In other patients, performing HRCT for ILD diagnosis as the first step is only realistic in selected populations with an increased risk for ILD or in whom medical history or clinical examinations revealed signs of ILD. Moreover, it must be considered that in some countries access to HRCT is limited. In these countries, complete pulmonary function testing is even more important to screen patients who are at risk for ILD and respiratory symptoms. A pattern of a restrictive ventilatory disorder and/or reduced diffusion capacity can provide initial diagnostic clues to the presence of ILD, which then lead to the initiation of further diagnostic steps. However, the interpretation of PFTs is influenced by inter-rater variability due to different guideline strategies over the years, lack of training or loss of expertise of the individual physicians.
In this study, we demonstrated for the first time that the use of AI for interpreting PFT increases the accuracy for detecting ILD significantly. Rather than relying solely on conventional thresholds and indices, the AI algorithms analyse the entire shape of the flow-volume curve and its relationship with other numbers including whether reduced DLCO or TLC indicates earlier signs of disease yet not apparent to usual interpretation.
Even without any prior knowledge and handling of AI, the accuracy for ILD identification is increased from 43% to 72%, a relative improvement of 68.5% from baseline. After a 6-month learning curve, accuracy was improved from 53% without AI to 75% with AI. As the diagnostic accuracy without AI also improved within this 6-month period, it can be hypothesised that AI does not only support the interpretation of an individual PFT but also trains the physicians in PFT interpretation and recognising lung function traits that are pointing to ILD. However, as the training data from the 6-month period were not incorporated into the AI model, no performance improvement was observed when comparing AI-supported PFT interpretation between phase one and phase 2.
In this study, ArtiQ.PFT software provided a correct diagnosis in 81%, which is similar to the result of a recent study that described a diagnostic accuracy of the ArtiQ.PFT software in 82% for PFT interpretation in various diseases.6 In both studies, the software outperformed the pulmonologists.
Similar results can be found not only in studies on software-assisted PFT interpretation but also in the use of AI in other areas of medicine. There are, for example, many trials related to AI use in CT interpretation for various diseases or AI-guided ECG interpretation that demonstrated a superiority of AI in comparison to physicians.15,17
However, these statements must be considered carefully and should not lead to a situation where PFTs, CT scans or electrocardiograms are only analysed by AI. A complete replacement of the physician by AI endangers the well-being of the patient and is not justifiable by personnel shortages. As the diagnostic accuracy of the software is not correct in 100% of the cases, the results of the examination must be interpreted by the physicians with consideration to the clinical history, physical examination and other diagnostic findings. The main goal of AI should therefore be to give the physician clues that must be interpreted in the context of the entire clinical presentation. If it is used in this way, AI-guided PFT interpretation may provide a timely diagnosis of ILD and allow the physicians to treat the patients earlier and to improve the patients’ outcomes.
One main limitation of this study is that it is exclusively an evaluation by questionnaires. The participating physicians only had a short medical history of the patient, but never investigated the patient themselves. In reality, a diagnosis is reached by a synergy of multiple factors, including expanded history, clinical examination and blood sampling. Having access to only incomplete clinical history and medical examination may also explain why the clinicians working together with AI did not improve on AI alone. In addition, it should be mentioned that because of the experimental design with repetitive questions, our physicians could have answered their diagnostic preferences with less attention. The real-life situation may, therefore, yield different outcomes. Although the results must be taken with caution, the massive increase in ILD recognition observed in this study confirms previous findings that clinicians improve their PFT interpretation when collaborating with AI software.7 The study participation was limited to physicians working in pulmonology departments, most of them were pulmonologists with more than 5 years of experience. Thus, the results of this study cannot be transferred to general practitioners who have less experience with PFT interpretation. However, it could be shown that non-specialists have significantly lower diagnostic accuracy without AI compared with specialists. When using AI, the increase in accuracy for ILD diagnosis was more pronounced in junior physicians compared with experienced pneumologists. This indicates that AI can help non-experts bridge the gap with experts. Similarly, Wang et al observed a training-dependent effect in recognising patterns of ventilatory impairment.18 The diagnostic accuracy of experienced vs less experienced clinicians, as well as clinicians from tertiary centres versus primary care centres, was inferior to that of a deep learning-based analytic model. However, the authors did not investigate how AI improves the interpretation of PFTs in physicians with varying levels of training. To the best of our knowledge, this is the first study to investigate the training-dependent effect of AI-based support in analysing PFTs.
A further limitation is that this study did not include patients with coexisting lung diseases, such as ILD and COPD. It is, therefore, unclear whether the benefit of AI-supported PFT interpretation is transferable to this patient population.
In conclusion, AI-guided PFT interpretation significantly improves the identification of ILD, providing physicians with an important opportunity to more accurately detect patients with ILD through body plethysmography, initiate earlier diagnostic steps, and potentially improve patient outcomes.
Supplementary material
Boehringer Ingelheim had no role in the analysis or interpretation of the results in this study.
Footnotes
Funding: This project was funded by Boehringer Ingelheim. Boehringer Ingelheim was given the opportunity to review the manuscript. Grant number is not applicable.
Provenance and peer review: Not commissioned; externally peer reviewed.
Patient consent for publication: Not applicable.
Ethics approval: This study involves human participants and Lung function data were collected within a trial in Belgium in 2017 that was approved by the ethics committee of the UZ Leuven centre (B322201732850). Participants gave informed consent to participate in the study before taking part. As no new patient data were collected within this study, no further ethics approval was required.
Data availability statement
All data relevant to the study are included in the article or uploaded as supplementary information.
References
- 1.Kaplan A, Cao H, FitzGerald JM, et al. Artificial Intelligence/Machine Learning in Respiratory Medicine and Potential Role in Asthma and COPD Diagnosis. J Allergy Clin Immunol Pract. 2021;9:2255–61. doi: 10.1016/j.jaip.2021.02.014. [DOI] [PubMed] [Google Scholar]
- 2.Krumpe P, Weigt G, Martinez N, et al. Computerized rapid analysis of pulmonary function test: use of a least mean squares correlation for interpretation of data. Comput Biol Med. 1982;12:295–307. doi: 10.1016/0010-4825(82)90033-6. [DOI] [PubMed] [Google Scholar]
- 3.Hankinson JL. Automated pulmonary function testing: interpretation and standardization. Ann Biomed Eng. 1981;9:633–43. doi: 10.1007/BF02364776. [DOI] [PubMed] [Google Scholar]
- 4.Stanojevic S, Kaminsky DA, Miller MR, et al. ERS/ATS technical standard on interpretive strategies for routine lung function tests. Eur Respir J. 2022;60:2101499. doi: 10.1183/13993003.01499-2021. [DOI] [PubMed] [Google Scholar]
- 5.Mac A, Xu T, Wu JKY, et al. Deep learning using multilayer perception improves the diagnostic acumen of spirometry: a single-centre Canadian study. BMJ Open Respir Res. 2022;9:e001396. doi: 10.1136/bmjresp-2022-001396. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Topalovic M, Das N, Burgel P-R, et al. Artificial intelligence outperforms pulmonologists in the interpretation of pulmonary function tests. Eur Respir J. 2019;53:1801660. doi: 10.1183/13993003.01660-2018. [DOI] [PubMed] [Google Scholar]
- 7.Das N, Happaerts S, Gyselinck I, et al. Collaboration between explainable artificial intelligence and pulmonologists improves the accuracy of pulmonary function test interpretation. Eur Respir J. 2023;61:2201720. doi: 10.1183/13993003.01720-2022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Spagnolo P, Ryerson CJ, Putman R, et al. Early diagnosis of fibrotic interstitial lung disease: challenges and opportunities. Lancet Respir Med. 2021;9:1065–76. doi: 10.1016/S2213-2600(21)00017-5. [DOI] [PubMed] [Google Scholar]
- 9.Global Initiative for Chronic Obstructive Lung Disease Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease (2024 report) 2024. [16-Dec-2024]. https://goldcopd.org/2024-gold-report/ Available. Accessed.
- 10.Global Initiative for Asthma Global strategy for asthma management and prevention (2024 update) 2024. [16-Dec-2024]. https://ginasthma.org/2024-report/ Available. Accessed.
- 11.Polverino E, Goeminne PC, McDonnell MJ, et al. European Respiratory Society guidelines for the management of adult bronchiectasis. Eur Respir J. 2017;50:1700629. doi: 10.1183/13993003.00629-2017. [DOI] [PubMed] [Google Scholar]
- 12.Humbert M, Kovacs G, Hoeper MM, et al. 2022 ESC/ERS Guidelines for the diagnosis and treatment of pulmonary hypertension. Eur Respir J. 2023;61:2200879. doi: 10.1183/13993003.00879-2022. [DOI] [PubMed] [Google Scholar]
- 13.Hewson T, McKeever TM, Gibson JE, et al. Timing of onset of symptoms in people with idiopathic pulmonary fibrosis. Thorax. 2018;73:683–5. doi: 10.1136/thoraxjnl-2017-210177. [DOI] [PubMed] [Google Scholar]
- 14.Hoffmann-Vold A-M, Maher TM, Philpot EE, et al. The identification and management of interstitial lung disease in systemic sclerosis: evidence-based European consensus statements. Lancet Rheumatol. 2020;2:e71–83. doi: 10.1016/S2665-9913(19)30144-4. [DOI] [PubMed] [Google Scholar]
- 15.Chen Y, Tian X, Fan K, et al. The Value of Artificial Intelligence Film Reading System Based on Deep Learning in the Diagnosis of Non-Small-Cell Lung Cancer and the Significance of Efficacy Monitoring: A Retrospective, Clinical, Nonrandomized, Controlled Study. Comput Math Methods Med. 2022;2022:2864170. doi: 10.1155/2022/2864170. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
- 16.Zhang Y, Jiang B, Zhang L, et al. Lung Nodule Detectability of Artificial Intelligence-assisted CT Image Reading in Lung Cancer Screening. Curr Med Imaging. 2022;18:327–34. doi: 10.2174/1573405617666210806125953. [DOI] [PubMed] [Google Scholar]
- 17.Noseworthy PA, Attia ZI, Behnken EM, et al. Artificial intelligence-guided screening for atrial fibrillation using electrocardiogram during sinus rhythm: a prospective non-randomised interventional trial. Lancet. 2022;400:1206–12. doi: 10.1016/S0140-6736(22)01637-3. [DOI] [PubMed] [Google Scholar]
- 18.Wang Y, Li Q, Chen W, et al. Deep Learning-Based Analytic Models Based on Flow-Volume Curves for Identifying Ventilatory Patterns. Front Physiol. 2022;13:824000. doi: 10.3389/fphys.2022.824000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Quanjer PH, Stanojevic S, Cole TJ, et al. Multi-ethnic reference values for spirometry for the 3-95-yr age range: the global lung function 2012 equations. Eur Respir J. 2012;40:1324–43. doi: 10.1183/09031936.00080312. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All data relevant to the study are included in the article or uploaded as supplementary information.