AI that reads lung cancer screening CT scans could uncover hidden COPD, but it isn't validated yet

By HCPLive · October 23, 2026.

Artificial intelligence analysis of lung cancer screening CT scans could help detect undiagnosed cases of COPD, but it still needs validation before it can be used routinely. That is the argument made by Jean Bourbeau, professor of respiratory medicine at McGill University, in an interview with HCPLive during the CHEST 2026 congress (Phoenix, October 18-21).

COPD (chronic obstructive pulmonary disease) is a progressive respiratory disease, usually linked to smoking. According to Bourbeau, smokers who meet lung cancer screening criteria are the same people at risk of COPD. Those CT scans generate a great deal of information that is not analyzed in depth today, and that is where AI would fit in: flagging at-risk patients so they can undergo spirometry (the lung function test that confirms the diagnosis).

The backdrop is the 2026 edition of the GOLD report, the international reference guideline on COPD. For the first time it includes a chapter (chapter 6) devoted to AI and emerging technologies, and it revises the section on screening and case finding. According to Bourbeau, GOLD 2026 addresses underdiagnosis and the 2027 edition will emphasize it even more. A manuscript under review will expand on the topic.

Bourbeau describes COPD as "the Alzheimer's of the lung": by the time patients reach the clinic, there is already a lot of disease. Earlier diagnosis would allow earlier treatment and help avoid emergency visits and hospitalizations. He stresses that many of these patients are not in an early stage; they were simply never diagnosed.

Bourbeau himself acknowledges that the technology is not ready for routine use, although he hopes it will be soon. The article provides no figures on sensitivity or accuracy, nor specific studies supporting the approach.

His team is also launching an AI program to identify the needs of patients leaving hospital after a COPD admission, with heart failure planned as the next step. It would show nurses each patient's most urgent daily tasks to prevent readmissions. It builds on years of research from the "Living Well with COPD" self-management program.

Our take: this is one of the most sensible applications of AI in medicine. It requires no new test or additional screening: it reuses an image already taken for another reason and extracts a second diagnosis from it. This is opportunistic detection, and it is usually where AI adds the most with the least friction.

But the promise should be separated from the evidence. What exists today is expert opinion and a guideline that acknowledges the limits, not results published in this material. That is why we think the design Bourbeau proposes is the right one: AI flags, the doctor confirms with spirometry. It keeps the clinician "behind" the technology, as he himself urges. The usual risks remain to be resolved: false positives overwhelming pulmonology clinics, and equity of access, since only those already enrolled in screening programs will benefit.

His most interesting warning is aimed at industry: academic groups with two decades of data and validated self-management methods should take part in the commercial development of AI, something that, he says, is lacking today. It is a reasonable criticism: without that clinical knowledge, commercial tools risk optimizing for what is measurable rather than what is useful.

Overall, in the short term we will see a slow and cautious transition, with validation, regulation and adaptation of workflows. In the long term, the direction is the one we advocate: medicine that detects disease while it is still treatable, almost automatically, and that brings closer a future in which chronic diseases are controlled or prevented. That the field's reference guideline already devotes a chapter to AI is a sign that this transition is beginning in an orderly way.