Episode #49 Lung Imaging 2.0: How AI is Shaping the Future with Dr. Rachel Eddy

Dr. Albert Rizzo: Welcome back to Lungcast, the monthly respiratory health podcast series from the American Lung Association and medical news site HCP Live. I'm your host, Dr. Albert Rizzo, chief medical officer of the American Lung Association.

Before we introduce our guest today, just a reminder that Lungcast is available on all your favorite streaming platforms. You can subscribe to @Lungcast on YouTube to get new and archived episodes, as well as interview segments and episode highlights as they come out. A link to the show page, as well as the ALA and HCP Live homepages, can be found in your episode description.

For those of you who prefer a podcast, you can subscribe to Lungcast on Spotify, Apple Podcasts, and your favorite listening platform. You can also register for more respiratory news and insights at both lung.org and HCPlive.com.

Now, let's go on to today's topic. By the late 20th century, applications of digital technology began to challenge the continued use of analog technology in medical imaging. The fields of radiography, fluoroscopy, and mammography made the transition by offering image acquisition and efficient storage and transmission methods known as PACS. These PACS, which stand for Picture Archiving and Communication Systems, streamlined healthcare operations while reducing costs associated with film-based processes.

Today, medical imaging technology continues to evolve. Applications of artificial intelligence have expanded beyond traditional radiology, aiding in pathology, dermatology, and even genomics. Imaging techniques—including 3D printing, molecular imaging, and functional MRI—are pushing the boundaries of diagnostics and treatment planning.

Today we're going to focus on artificial intelligence and its role in chest imaging, and for that, my guest today is Dr. Rachel Eddy.

Dr. Albert Rizzo: Dr. Eddy holds a joint appointment as an assistant professor in the Department of Pediatrics at the University of British Columbia and British Columbia Children's Hospital in Radiology. She’s an investigator with the University of British Columbia Center for Heart and Lung Innovation and British Columbia Children’s Hospital Research Institute. She completed her undergraduate degree in electrical and biomedical engineering at McMaster University and a PhD in medical biophysics at Western University.

Dr. Eddy's research has focused on developing and applying novel pulmonary imaging and data science tools to provide a deep understanding of chronic lung disease and novel inhalation lung exposures, as well as ways to use imaging to guide lung disease interventions. She sounds like an excellent guest for today's topic, so thank you for joining me today, Dr. Eddy.

Dr. Rachel Eddy: Yes, thank you very much for having me here.

Dr. Albert Rizzo: Artificial intelligence is an umbrella term that incorporates various algorithms, including machine learning, deep learning, and natural language processing. Could you please give our listeners a brief tutorial on these different terms, and comment on the more recent use of the term “generative” when it comes to AI and imaging?

Dr. Rachel Eddy: Sure, such a big topic! I’ll try to summarize briefly. We can think of AI as a large, all-encompassing circle. AI is generally described as the ability of a machine to learn automatically and display intelligent, human-like behavior. It relies on big datasets to develop its function in a consistent fashion and generate reliable results.

Within AI, machine learning is a smaller circle. It involves machines automatically learning patterns in data and using those patterns to make decisions about new data. Machine learning maps input to output using math and statistics, creating explainable relationships that we can trace.

Deep learning is a smaller circle within machine learning. It uses more complex algorithms, typically called deep neural networks, to mimic the human brain. Deep learning requires very large datasets to learn accurate mappings. Unlike machine learning, deep learning is often called a “black box” because it’s harder to see exactly how the machine arrived at its result.

Both machine learning and deep learning have been applied extensively in lung imaging. They help with disease detection, classification (for example, determining if a lung nodule is cancerous), measurements, and predicting patient outcomes. Generative AI, the latest development, can create new content like text, images, or audio. In medical imaging, generative AI can produce radiology-like reports from images or even generate 3D CT images from 2D chest X-rays.

Dr. Albert Rizzo: That’s a great overview. Let’s go more deeply into your work. I understand you’ve applied AI to assess airway and mucus changes in asthma. Can you share what your work has taught us and how it translates to conditions like COPD or asthma?

Dr. Rachel Eddy: Yes. My work has used AI algorithms to derive measurements from CTs of the lungs, especially the airways. Asthma has long been understood as a variable lung disease, with symptoms that come and go and are reversible with inhaled therapies. We’ve shown that airway abnormalities in asthma are very focal—like a unique fingerprint for each person.

We conducted a longitudinal study and found these focal airway abnormalities persisted over at least six and a half years. These imaging measurements could predict future loss of reversibility with bronchodilators. This connects to COPD, which is defined as a fixed airflow obstruction. Some patients with asthma have airway measurements similar to COPD and may progress toward that phenotype.

Additionally, we’ve shown that airway measurements and mucus plugging respond to different asthma treatments, providing quantitative imaging tools to assess treatment response. Similar approaches in COPD have shown that patients with more mucus plugs have a higher risk of mortality, highlighting potential treatment targets.

Dr. Albert Rizzo: Outside of airway diseases, you’re also working on imaging biomarkers in interstitial lung disease (ILD). How is AI transforming the diagnostic process here?

Dr. Rachel Eddy: In ILD, CT is critical for diagnosis. Expert radiologists currently evaluate CTs for patterns like reticulation, traction bronchiectasis, and honeycombing, often providing crude estimates of fibrosis burden. AI can precisely quantify total fibrosis and individual patterns, predict mortality risk, evaluate treatment response, and even serve as an inclusion criterion in clinical trials to enrich study populations. Automated algorithms can also detect early interstitial lung abnormalities before overt changes are visible on CT.

Dr. Albert Rizzo: The American Lung Association is conducting the Lung Health Cohort Study, tracking Millennials to identify early lung changes. How do you see AI applied in this study?

Dr. Rachel Eddy: AI can automatically measure subtle abnormalities in lung structure, like airway wall thickening and mucus plugging, even before disease is clinically diagnosed. AI can also learn patterns linking imaging with blood biomarkers or environmental factors, identifying novel predictors of future disease. Supervised learning trains models to measure known features, while unsupervised learning lets models identify unknown predictive features in the data.

Dr. Albert Rizzo: Can you explain the ancillary studies involving deep lung phenotyping with CT and hyperpolarized Xenon MRI?

Dr. Rachel Eddy: The CT-based synapse measurement evaluates the ratio of airway size to lung volume, providing insight into lung structure as a risk factor for COPD. Xenon MRI complements CT by measuring lung function at a microstructural level. Patients inhale Xenon gas, and we image how it distributes and exchanges in the lungs. This detects subtle early changes in small airways that may precede COPD or other lung disease.

Dr. Albert Rizzo: From a technical and clinical perspective, where do you see AI in lung imaging evolving over the next five years?

Dr. Rachel Eddy: Technically, advances in computing are enabling the use of full 3D datasets, capturing regional heterogeneity in lungs. Clinically, AI can support physicians by reviewing imaging, identifying patient phenotypes, and aiding prognostication. The key is balance—AI should support, not replace, expert clinicians. Algorithms need to be developed with large, diverse datasets for generalizable results.

Dr. Albert Rizzo: Thank you for your time today, Dr. Eddy, and for sharing your expertise with our listeners.

To our listeners, if you want to hear more about respiratory health, be sure to subscribe and rate Lungcast on your preferred platform, and visit lung.org and HCP.com for more news and resources.

Until next time, I’m Dr. Albert Rizzo, reminding you that if you can’t breathe, nothing else matters.