Artificial intelligence (AI) is undoubtedly dominating the contemporary cultural ethos, with pervasive, exciting, and anxiety-inducing implications for humanity. Not surprisingly, AI applications in health care are multiplying, and AI-focused research is gaining traction across specialties, including pulmonary, critical care, and sleep medicine.
“AI is being widely used across multiple domains in health care [and] across the continuum of patient care—from when the patient first presents with symptoms at the clinic all the way up to when they’re undergoing a procedure or posttreatment evaluation,” said William Healy, MD, FCCP, a pulmonary, critical care, and sleep medicine physician at Wellstar MCG Health Medical Center in Georgia.
AI terminology primer
As CHEST President-Elect Neil Freedman, MD, FCCP, a pulmonary, critical care, and sleep medicine physician at Endeavor Health in Illinois, put it, “AI means a lot of things to a lot of people.” Thus, clarity and specificity are needed regarding terminology when discussing AI applications.
AI is an umbrella term referring to the use of coded programs or algorithms for tasks requiring objective reasoning and understanding.1 Most AI instances or programs in medicine rely on machine learning models, which identify patterns in training datasets that can then be applied to analyze new or model-naïve scenarios. Deep learning models—a subset of machine learning—involve larger training datasets, use neural networks with hidden layers, and can independently extract features from new data.2
The increasing significance of AI technology is exemplified by the AI Glossary, an educational resource from the US Food and Drug Administration (FDA). Additionally, medical professional societies and public health organizations, including the American Medical Association and the World Health Organization, are increasingly issuing recommendations or standards regarding AI in health care.3–4
Here, we will examine a sampling of AI use cases across the continuum of pulmonary medicine—from education and evaluation of trainees to the first encounter of a patient with their health care provider and further along the care pipeline in sleep medicine and lung cancer care. More in-depth looks at various AI innovations in pulmonary, critical care, and sleep medicine will be explored in future CHEST Physician issues.
Tools for educators, fellowship directors
Alison Whelan, MD, Chief Academic Officer of the Association of American Medical Colleges (AAMC), recently noted that the era of AI-enabled transformations in education and training for US clinicians is just beginning.5
Bhavinkumar Dalal, MD, MBBS, FCCP, Professor of Pulmonary, Critical Care, and Sleep Medicine at Corewell Health in Michigan, agreed. “Broadly, AI applications are now being used by clinician educators for curriculum development, in learner assessments, and even for feedback and evaluation,” he said.
Dr. Dalal illustrated how AI applications are being integrated into the fellow recruitment cycle at his institution. AI tools can be used to calculate z scores for the typical 500-strong applicant pool for the six fellowship positions per cycle. They can also summarize applications and interview feedback for short-listed applicants before interviews with faculty members. The algorithm can then apply predefined rubrics to the qualitative summaries, folding this rubric into the z score calculations.
While several general AI tools, such as Copilot and ChatGPT, have been used in this context, Dr. Dalal said that his group is planning a formal study to standardize the use of AI aids and assess their impact on recruitment-related parameters.
Dr. Dalal has also used AI for developing a situational clinical judgement assessment focused on pulmonary and critical care. He said that the AI-generated test, composed of 20 questions, was a reasonable starting point. He subsequently prompted and refined the test through iterations to generate the final version. A previous study showed that AI-generated assessments for mechanical ventilator competency were comparable with human expert-created tests.6
Dr. Dalal and colleagues are also studying the correlation between candidate ranking based on an AI-facilitated assessment and faculty-generated ranking. Another AI-enabled curriculum for interstitial lung disease, based on Kern’s six-step model, is in development.
“The AI-aided approach provided rubrics and frameworks for all six steps, which I then had to fine-tune and modify,” Dr. Dalal said.
Many medical schools are in the process of implementing AI curriculums for trainee clinicians. The AAMC, for instance, has an educational series on AI that provides guidance on critical concepts and practical strategies. Likewise, the Association of Pulmonary and Critical Care Medicine Program Directors offers hands-on AI workshops for clinician educators.
AI and the patient-provider interface
Across the spectrum of pulmonary specialists, there is consensus that the most prevalent use of AI in contemporary real-world practice centers around reducing administrative workload for providers, a critical and welcome change with the potential to reduce clinician burnout.
“In general practice, for instance, Ambient AI can be used to help physicians with coding and documentation, freeing up more time to spend with patients,” Dr. Freedman said. “Other AI-enabled tools [such as DAX Copilot in Epic and AI Scribe] are already in use for a range of administrative tasks and to improve the patient’s care experience, such as for answering patient calls, fulfilling medication refill requests, and handling insurance precertification and reimbursements.”
Moving forward in sleep medicine
As noted by Ritwick Agrawal, MD, MS, FCCP, Director of Sleep Medicine at Northwell Health in New York, sleep medicine has traditionally been a data-rich field, typified by the seven to eight hours of data from an individual multichannel sleep study. Even before the advent of AI tools, sleep medicine specialists were pioneers in technology and digital tools, he said.
“AI tools are now being deployed to analyze multichannel sleep study results, along with the electronic medical record, to identify patients who may be at a high risk for sleep disorders and provide screening recommendations. Some of these AI tools are replacing or supplementing manual sleep study scoring,” Dr. Agrawal said. “Neural network models for extracting sleep parameters from polysomnography data collected with in-home sleep diagnostic devices are being developed.”
However, most AI applications in sleep medicine are still investigational or in the research setting. For example, Dr. Agrawal and colleagues are developing an AI model for assessing the STOP-Bang score, a patient-driven screening tool for OSA.
Such a tool would help address a significant unmet need, particularly in primary care: identifying patients who may be at risk of sleep disorders and guiding referrals for follow-up or testing.
Drs. Freedman and Agrawal said AI tools may also help address other challenges in sleep medicine, such as predicting patient tolerability and adherence to CPAP therapy or patient suitability for different treatments (eg, hypoglossal nerve stimulation or surgical approaches). The availability of validated AI tools tailored for these needs not only may improve personalized patient care and outcomes but also may help to reduce health care costs.
Unmet needs in lung cancer
William Mayfield, MD, Medical Director of Lung Cancer Screening and Incidental Nodule Programs at Wellstar Health System in Georgia, is a principal investigator in the Sybil Implementation Consortium, an alliance focused on developing and implementing the Sybil AI model for predicting risk of lung cancer.7
Dr. Mayfield said AI has several applications in the general pathways of lung cancer, such as suspicious nodule detection, nodule characterization, and—with the development of Sybil—the potential for predicting future lung cancer risk.
Several commercially available AI tools, such as Riverain Technologies and QureE AI, can independently review CT scans, detect lung nodules, and then flag them for the radiologist. These tools may be particularly useful in settings where imaging data are reviewed by radiologists with limited experience in thoracic imaging, at high-volume centers, or in other resource-limited settings.
Other AI tools can parse radiology reports, highlight the term “nodule,” and then route that report to reviewers or nurse navigators for follow-up.
“These AI tools can level the playing field for everyone,” Dr. Mayfield said. “With AI-assisted identification of suspicious nodules warranting further evaluation, the radiologist’s attention is focused, and they can refer patients and tailor next steps.”
Another AI-enabled lung cancer prediction tool, Optellum’s Lung Cancer Prediction score, is the world’s first FDA-approved imaging AI/radiomics-based digital biomarker. This tool ascribes a lung cancer risk score to nodules specifically delineated by a clinician in a CT scan and designated for assessment.
The Sybil risk prediction tool was developed by researchers at the Jameel Clinic at the Massachusetts Institute of Technology. Led by Regina Barzilay, School of Engineering Distinguished Professor of AI and Health—who also spearheaded Mirai, a mammography-based risk prediction model for breast cancer—Sybil was validated in diverse cohorts, demonstrating robust and consistent performance in individualized risk prediction for lung cancer development within six years from an index CT scan image, independent of clinical or medical historical information.8–10 Additional validation and prospective studies incorporating Sybil are planned.
“Currently, Sybil is an investigational/research tool, and many aspects must be addressed before it can be truly ready for routine clinical application and inform patient care decisions,” Dr. Mayfield said.
Evolving uses, unresolved issues, and challenges
AI evolution in medical practice is happening rapidly, Dr. Mayfield said, “and accelerating exponentially at a rate faster than any other technology in the history of mankind.”
Given this rapid expansion, growing pains are to be expected.
During this evolution, clinicians will need to understand how to work with—not be replaced by—AI tools, Dr. Freedman said.
“We need to understand the key shortcomings of these emerging tools and the governance and regulatory structures in our practices and institutions that need to be in place to optimize their use appropriately and in compliance with exiting health care regulations and privacy laws,” he said.
Dr. Healy added, “Many current AI tools have narrow applications or can only be deployed in well-funded, adequately resourced care centers. Broadly applicable AI tools—prospectively validated in large datasets representing diverse populations and accessible across practice settings, including rural or resource-limited care settings—are needed.”
Professional medical organizations or research consortia can help assemble large datasets and provide guidance and recommendations for systematic development of AI tools, both to address research priorities and for use in clinical practice, he said.
“AI tools are certainly here to stay,” Dr. Healy said, echoing the perspectives of his peers.
This article was originally published in the Winter 2025 issue of CHEST Physician.
References
1. National Academies of Sciences, Engineering, and Medicine; Health and Medicine Division; Board on Global Health; Global Forum on Innovation in Health Professional Education; Forstag EH, Cuff PA. Glossary of Terms. In: Artificial Intelligence in Health Professions Education: Proceedings of a Workshop. National Academies Press (US); 2023. Accessed October 1, 2025.
2. Brower D, Sengupta S, Bhatt AN, et al. Artificial intelligence in interventional pulmonology. Ther Adv Pulm Crit Care Med. 2025;20:29768675251353390. doi:10.1177/29768675251353390
3. American Medical Association. Augmented Intelligence Development, Deployment, and Use in Health Care. November 2024. Accessed September 30, 2025.
4. World Health Organization. WHO releases AI ethics and governance guidance for large multi-modal models. January 18, 2024. Accessed October 2, 2025.
5. Boyle P. AI in medical education: 5 ways schools are employing new tools. Association of American Medical Colleges. February 27, 2025. Accessed October 2, 2025.
6. Safadi S, Amirahmadi R, Tlimat A, et al. Quality of human expert vs large language model-generated multiple-choice questions in the field of mechanical ventilation. Chest. 2025:S0012-3692(25)00837-2. doi:10.1016/j.chest.2025.07.005
7. Mikhael PG, Wohlwend J, Yala A, et al. Sybil: a validated deep learning model to predict future lung cancer risk from a single low-dose chest computed tomography. J Clin Oncol. 2023;41(12):2191-2200. doi:10.1200/JCO.22.01345
8. Behrman H. Sybil AI Model Demonstrates Strong Performance in Predicting Lung Cancer Risk with Minimal Bias. ILCN.org (ILCN/WCLC). September 7, 2025. Accessed October 2, 2025.
9. Lee JH, Chae KJ, Lu MT, et al. External testing of a deep learning model for lung cancer risk from low-dose chest CT. Radiology. 2025;316(2):e243393. doi:10.1148/radiol.243393
10. Li H, Salehjahromi M, Godoy MCB, et al. Lung cancer risk prediction in patients with persistent pulmonary nodules using the Brock model and Sybil model. Cancers (Basel). 2025;17(9):1499. doi:10.3390/cancers17091499
11. Suleyman M, Bhaskar M. Chapter 1: Containment is Not Possible. In: The Coming Wave. Penguin Random House, LLC.
