The current landscape of lung radiology and pathology integrates artificial intelligence (AI) and computational models, and advanced functional imaging modalities, as well as leveraging multimodal, multidisciplinary approaches.
“It’s a very exciting time for AI, not just for COPD but for health care in general. AI will likely be applied in nearly every area of chronic lung disease management, from drug discovery to phenotyping, from diagnosis to treatment,” said Meilan King Han, MD, Professor of Medicine and Chief of the Division of Pulmonary and Critical Care at the University of Michigan and a member of the Fleischner Society and the Global Initiative for Chronic Obstructive Lung Disease (GOLD) Scientific Committee.
Lung imaging and digital pathology
Anant Madabhushi, PhD, Professor of Biomedical Engineering at the Georgia Institute of Technology and Emory University and Executive Director of the Emory Empathetic AI for Health Institute, highlighted the breadth of AI applications in pulmonary medicine, including lung cancer screening, treatment response prediction, prognostication, infectious respiratory disease detection, and fibrotic lung disease management.1
“Lung cancer screening is an area of significant interest, with many AI-enabled tools, such as Optellum’s Virtual Nodule Clinic, gaining US Food and Drug Administration clearance.2 Risk-stratification tools are also emerging, such as the Sybil model for predicting future lung cancer risk,” Dr. Madabhushi said. There are even AI models that can predict treatment response.
AI-enabled prediction of tumor mutational profiles is another emerging area.3 “AI models are also being developed in the infectious respiratory disease space, such as for querying chest X-ray images to predict the presence of TB and to differentiate TB from cancer,” Dr. Madabhushi said, referring to the qure.ai’s qXR and qTrack AI-powered models.4
“Such AI algorithms are going to be very important, particularly in the Global South where TB has high prevalence,” he said, adding that models evaluating both chest CT scan and X-ray images may be more beneficial for opportunistic applications and in resource-limited settings. For instance, Dr. Madabhushi’s group developed an AI-based tool to diagnose HIV status in patients with TB using opportunistic X-ray images obtained from patients in Tanzania and Uganda.5
Addressing unmet needs and challenges
The significance of AI in contemporary pulmonary medicine is perhaps exemplified by the inclusion of an entirely new chapter in the GOLD 2026 report, “Artificial Intelligence and Emerging Technologies in COPD.”6
The GOLD Scientific Committee is also drafting another manuscript focused on the role of AI-enabled approaches in COPD management, Dr. Han said, adding that AI-enabled tools provide opportunities to address the underdiagnosis of chronic respiratory diseases (CRDs).
CRDs remain a persistent public health challenge. COPD is a leading cause of morbidity and mortality, accounting for 5% of all deaths globally, while asthma has the highest prevalence worldwide.7–8 COPD and asthma underdiagnosis are associated with increased burdens on health care systems and a reduced health-related quality of life.9–11
Young Juhn, MD, MPH, Professor of Pediatrics, Co-Director of the Center for Clinical and Translational Science, and Director of the AI Program Promoting Adolescent and Childhood Health (APPROACH) initiative at Mayo Clinic Rochester, highlighted diagnostic challenges in asthma.
“Delayed diagnosis of asthma is a huge issue. In survey-based studies in the US and Europe, for instance, only around 20% of young children who experience recurrent weekly asthma-like symptoms receive an asthma diagnosis, and of those, only 10% receive inhaled corticosteroid therapy,” Dr. Juhn said. “The lung function trajectory that sets patients on the course to a diagnosis of a chronic lung disease like asthma or COPD at age 50 or beyond often starts in early childhood.”
To improve early COPD detection, Dr. Han is spearheading the development of an AI model that leverages electronic health records (EHRs) to identify high-risk patients. AI-based COPD risk scores—such as in the prospective DYNAMIC-AI UK trial—demonstrate potential for advancing population-level COPD care.12
Deep learning models for diagnosing COPD using opportunistic CT scan imaging data, such as those obtained during lung cancer screening, have also been developed by several groups, including Dr. Han and his colleagues, who expanded on prior research where quantitative emphysema thresholds helped endotype COPD.13
A model for AI integration
“At Mayo, we have been working on several AI-powered tools within the chronic lung disease space,” Dr. Juhn said.
In addition to using AI-based approaches to reduce the administrative, physical, and cognitive burdens for clinicians, Dr. Juhn and colleagues developed the Asthma Guidance and Prediction System (A-GPS). This AI-powered clinical decision support system for asthma provides a single-screen summary of relevant patient information and integrates natural language processing (NLP) and machine-learning models to predict risk of asthma exacerbations. It also offers explanations for the risk and suggestions for improving care quality.
A-GPS combines multiple AI models, including the NLP-Predetermined Asthma Criteria (PAC) and NLP-Asthma Predictive Index (API), to help identify patients who meet diagnostic criteria but have never received an asthma diagnosis.14–16
“One of the significant advantages with our clinically validated AI models, like NLP-PAC and NLP-API, is that we can identify a subgroup of children with asthma as young as 3 years old who are at risk of acute respiratory infection (eg, pneumonia) or exacerbations,” Dr. Juhn said.
NLP models leverage the totality of free-form textual information in EHRs to predict the risk of asthma exacerbation and acute respiratory infections rather than relying solely on specific imaging studies or individual clinical variables.
A-GPS, which integrates one of the most mature models validated in many clinical trials, also reduced the time required to gather information and make clinical decisions by about 70%, Dr. Juhn said.
An advantage of the A-GPS platform is the design (SMART on FHIR platform), which facilitates seamless ad hoc integration of new or revised AI models. This includes remote patient-monitoring devices and the point-of-care delivery of essential, real-time contextual clinical information.
The asthma-specific A-GPS framework can also be adapted to patient management across specialties, Dr. Juhn noted. “We are currently evaluating the A-GPS platform schema to extend for COPD, gastroenterology, neuro-oncology care, and transplant medicine,” he said.
Moving models into clinical practice
Along with the optimism surrounding the potential of AI-enabled tools and the critical role of multi-institutional clinical validation of AI algorithms, several recurring themes emerged, highlighting caveats and challenges.
Dr. Madabhushi said that his group focuses not only on validating AI-powered models in prospective clinical trials but also retrospectively in completed clinical trials to generate higher (Level 1B) levels of evidence—a key consideration when drafting clinical practice guidelines.
“While over a thousand technologies have FDA approval, with a multitude also currently being reimbursed, few have undergone rigorous clinical validation in adequate datasets representing diverse populations. The lack of high-level evidence is one reason for the AI hype but relatively low clinical adoption and buy-in/comfort for clinicians,” Dr. Madabhushi said.
Dr. Juhn emphasized the importance of intentionality in the design and development of clinical AI tools.
“To people holding AI hammers, every clinical problem may look like a nail,” he said. “But we need to be selective and recognize that AI models that leverage conventional methods, such as logistic regression-based prognostic or predictive tools, still offer tremendous resources.”
He added that early stakeholder engagement during the development of clinical AI models is critical to direct resources toward key, unresolved challenges, ensuring the representation of diverse perspectives and fostering model acceptability.
Drs. Madabhushi and Juhn are strong proponents of explainable, interpretable AI models.
“One of the reasons we were able to embed our AI algorithm in a prospective validation study in lung cancer was the explainable feature of the tortuosity of tumor vasculature as a digital imaging biomarker predictive of response and survival,” Dr. Madabhushi said.17–18
“Rule-based NLP models are readily explainable and can therefore be more accessible, understandable, and trustworthy for clinicians and often perform as well as a large-language model,” Dr. Juhn added. “We will need to focus on improving our understanding of how to establish a trustworthy human-AI partnership pertaining to when to trust humans and when to trust AI.”
Dr. Han discussed the challenge of legal and regulatory frameworks. “While conventional and standard quantitative algorithms, particularly CT-focused methods, are more readily accepted, the pathway to regulatory approval and qualification is more challenging for other types of AI-based algorithms. However, this may change in the future.”
Another recurring theme is leveraging multimodal AI models to improve patient care.
AI models that combine unsupervised elements and quantitative data seem to yield the best clinical performance, Dr. Han said. He added that multimodal AI models—such as MERLIN, trained on both EHR and CT data, and Pillar-0, a foundational chest CT scan-based model—represent the “next frontier.”19–20
“There is so much information,” Dr. Madabhushi said. “We owe it to our patients to take the totality of available data and be able to apply AI to truly and compellingly provide the most informed prognostic or treatment response assessments.”
References
1. Viswanathan VS, Toro P, Corredor G, et al. The state of the art for artificial intelligence in lung digital pathology. J Pathol. 2022;257(4):413-429. doi:10.1002/path.5966
2. Optellum Inc. Providers overview (USA). https://optellum.com/providers-overview-usa/
3. Bera K, Braman N, Gupta A, et al. Predicting cancer outcomes with radiomics and artificial intelligence in radiology. Nat Rev Clin Oncol. 2022;19(2):132-146. doi:10.1038/s41571-021-00560-7
4. Qure.ai. https://www.qure.ai/
5. Cherezov D, Dam T, Najjingo I, et al. Opportunistic use of artificial intelligence with X-ray imaging for diagnosis of HIV status in tuberculosis patients in Uganda and Tanzania. PLoS Digit Health. 2025;4(9):e0000988. Published Sept. 2, 2025. doi:10.1371/journal.pdig.0000988
6. Global Initiative for Chronic Obstructive Lung Disease (GOLD). Key Changes in the Global Strategy for Prevention, Diagnosis, and Management of COPD: GOLD 2026. November 10, 2025. https://goldcopd.org/wp-content/uploads/2025/11/KEY-CHANGES-GOLD-2026-10Nov2025.pdf
7. World Health Organization. Chronic obstructive pulmonary disease (COPD) fact sheet. https://www.who.int/news-room/fact-sheets/detail/chronic-obstructive-pulmonary-disease-(copd)
8. Cao Z, He L, Luo Y, et al. Burden of chronic respiratory diseases and their attributable risk factors in 204 countries and territories, 1990–2021: Results from the global burden of disease study 2021. Chin Med J Pulm Crit Care Med. 2025;3(2):100-110. doi:10.1016/j.pccm.2025.05.005
9. Gerstein E, Bierbrier J, Whitmore GA, et al. Impact of undiagnosed chronic obstructive pulmonary disease and asthma on symptoms, quality of life, healthcare use, and work productivity. Am J Respir Crit Care Med. 2023;208(12):1271-1282. doi:10.1164/rccm.202307-1264OC
10. Alhabeeb FF, Whitmore GA, Vandemheen KL, et al. Disease burden in individuals with symptomatic undiagnosed asthma or COPD. Respir Med. 2022;200:106917. doi:10.1016/j.rmed.2022.106917
11. Lin CH, Cheng SL, Chen CZ, et al. Current progress of COPD early detection: key points and novel strategies. Int J Chron Obstruct Pulmon Dis. 2023;18:1511-1524. doi:10.2147/COPD.S413969
12. Taylor A, Hickey L, Walker E, et al.M37 Deploying live AI-based risk prediction models for use in a COPD MDT: acceptability, feasibility and utility data from the DYNAMIC-AI clinical trial. Thorax.2024;79:A292-A293. doi:10.1136/thorax-2024-BTSabstracts.465
13. Labaki WW, Xia M, Murray S, et al. Quantitative emphysema on low-dose CT imaging of the chest and risk of lung cancer and airflow obstruction: an analysis of the national lung screening trial. Chest. 2021;159(5):1812-1820. doi:10.1016/j.chest.2020.12.004
14. Wi CI, Sohn S, Ali M, et al. Natural language processing for asthma ascertainment in different practice settings. J Allergy Clin Immunol Pract. 2018;6(1):126-131. doi:10.1016/j.jaip.2017.04.041
15. Juhn YJ, Wi CI, Ryu E, et al. Artificial intelligence biomarker detects high-risk childhood asthma subgroup for respiratory infections and exacerbations. J Allergy Clin Immunol. 2025;156(6):1547-1555.e4. doi:10.1016/j.jaci.2025.07.031
16. Sohn S, Wi CI, Wu ST, et al. Ascertainment of asthma prognosis using natural language processing from electronic medical records. J Allergy Clin Immunol. 2018;141(6):2292-2294.e3. doi:10.1016/j.jaci.2017.12.1003
17. Vaidya P, Khorrami M, Bera K, et al. Computationally integrating radiology and pathology image features for predicting treatment benefit and outcome in lung cancer. NPJ Precis Oncol. 2025;9(1):161. Published June 4, 2025. doi:10.1038/s41698-025-00939-0
18. Alilou M, Khorrami M, Prasanna P, et al. A tumor vasculature-based imaging biomarker for predicting response and survival in patients with lung cancer treated with checkpoint inhibitors. Sci Adv. 2022;8(47):eabq4609. doi:10.1126/sciadv.abq4609
19. Blankemeier L, Cohen JP, Kumar A, et al. Merlin: A vision language foundation model for 3D computed tomography. Preprint. Res Sq. 2024;rs.3.rs-4546309. Published June 28, 2024. doi:10.21203/rs.3.rs-4546309/v1
20. Agrawal KK, Liu L, Lian L, et al. Pillar-0: A new frontier for radiology foundation models. arXiv. 2025:2511.17803. doi:10.48550/arXiv.2511.17803
