Breathing Room
By Beth W. Orenstein
Radiology Today
Vol. 26 No. 3 P. 14

AI and CT are augmenting COPD assessment.

In 2021, the most recent year for which statistics are available, COPD caused 3.5 million deaths worldwide, making it the fourth leading cause of death, according to the World Health Organization (WHO). With the number of cases on the rise, WHO predicts that by 2030, COPD will become the third leading cause of death worldwide.

The gold standard for diagnosing COPD has long been the pulmonary function test (PFT) spirometry. The spirometry test measures how much air patients can breathe in and out and how quickly they can do so. It uses a device known as a spirometer with a mouthpiece that patients must breathe into after taking a deep breath. Usually, the test is performed three times to make sure the results are consistent. Significantly, spirometry is a global assessment of the lungs and, therefore, is not a sensitive assessment of lung function. “Spirometry is probably the oldest and sort of original pulmonary function test,” says David Westenkirchner, RRT, director of clinical products for 4DMedical, which produces software and hardware products that enable noninvasive, quantitative imaging and analysis of lung function. A problem with spirometry is that it is uncomfortable and time consuming, says Greg Mogel, MD, FACR, CMO of 4DMedical.

It can be difficult for patients, particularly the elderly and those with impaired lung function, to repeatedly hold their breath and breathe in and out, as the spirometry test requires, Westenkirchner says.

Over the last decade-plus, new technologies that analyze CT lung scans have begun to supplement, if not replace, spirometry, Mogel says. CT lung scans are also being used in the treatment and care of COPD patients, he says.

Advanced technologies such as 4D Medical’s CT LVAS (Lung Ventilation Analysis) provide clinicians with detailed information about regional lung function, using CT images from existing hospital equipment. CT LVAS quantifies regional lung ventilation and ventilation heterogeneity in a report that radiologists can easily interpret. It includes 4D visualizations and enables both lung structure and function to be assessed in one procedure.

4DMedical’s Lung Density Analysis, Functional (LDAf) is designed to support physicians in diagnosis, care planning, and follow-up for patients with conditions where air-trapping may be present, such as COPD. LDAf software runs automatically on paired inspiratory/expiratory chest CT series without any user intervention. LDAf provides a complete mapping of normal lung, air-trapping, and areas of persistent low density in a combined image to help visualize the components of COPD, while its Lung Density Analysis, Inspiration (LDAi) algorithm aids in quantification and visualization of emphysema, a type of COPD, in low-dose CT scans. LDAi also produces personalized reports for face-to-face patient education on the conditions of their lungs and for use in smoking cessation programs. COPD is usually caused by cigarette smoking and accounts for eight out of 10 COPD deaths, according to the CDC.

A Deep Breath
Another company, Riverain Technologies, also has FDA-cleared tools that aid clinicians in diagnosing lung diseases, including COPD. Since 2021, Riverain has partnered with 4DMedical (through its acquisition of Imbio, a medical imaging AI provider for chronic lung and cardiothoracic diseases). Like Mogel and Westenkirchner, Jason Knapp, chief technology officer at Riverain, believes meaningful information can be extracted from CT scans that can aid COPD diagnosis and treatment, thanks largely to AI. “It’s not that you necessarily look at a CT scan and say, ‘This is COPD,’” Knapp says. What AI can do, though, is identify biomarkers that can in turn help identify the disease. For example, the location and distribution of so-called mucus plugs throughout the airways “are strongly correlated not just with whether you’re looking at asthma or COPD but also the state of the disease itself and how a patient might respond to this or that therapy,” he says.

These lung function imaging software products have been available since about 2016, Knapp says, but they are being used more and more as physicians have become more comfortable with AI and CT scanners have become more sensitive, with less variation in their images. He says it’s likely AI and CT imaging are taking a larger role in diagnosing and treating COPD because the computing ability of the technology “is where it needs to be.” However, he says, it is possible that “computational infrastructure” has, at least in part, been “an enabler.”

“Is the field taking advantage of the technology or is the technology pushing this change? I don’t know,” Knapp says.

Research on the use of CT and its role in COPD is ongoing and gaining steam. One project that is of interest is underway at San Diego State University in California. Similar to spirometry, two CT image acquisitions of the lungs are typically required to aid in COPD diagnosis and treatment: one inspiratory and one expiratory.

However, Kyle A. Hasenstab, PhD, an assistant professor of statistics and data science at San Diego State University, says this type of protocol is not clinically standard across institutions. Expiratory CT protocols may not be feasible at many institutions because they don’t have the means to train technologists to acquire the images or radiologists to interpret them, Hasenstab says. Also, as with spirometry, elderly patients with impaired lung function often struggle to hold their breath, which is required during exhalation image acquisition.

In December 2024, Hasenstab and colleagues published a study in Radiology: Cardiothoracic Imaging that found using one inhalation lung CT scan and a deep learning model could accurately diagnose and stage COPD. The researchers hypothesized that a single inhalation CT acquisition combined with a convolutional neural network (CNN) and clinical data could be sufficient for diagnosing and staging COPD. (A CNN is a type of neural network that uses deep learning to analyze and classify images.)

“Given that an expiratory acquisition is not clinically standard, we sought to determine whether a single inspiratory acquisition could instead be used for COPD diagnosis and staging, without the use of an expiratory acquisition,” Hasenstab says.

Saw It on CNN
The CNN was developed on the COPDGene Phase 1 cohort, which included 10,198 current and former smokers enrolled between 2007 and 2011. The cohort was part of a multi-institutional study funded by the National Heart, Lung, and Blood Institute to identify the genetics of COPD. Hasenstab notes that because the cohort may not represent the variation of data observed for clinical care, reduced performance may be observed when applying the algorithm to images with other pathologies.

To conduct the retrospective study, the researchers acquired inhalation and exhalation lung CT images and spirometry data from nearly 9000 patients. The average age of the patients in the study, all of whom had a history of smoking, was 59 years. They trained CNNs to predict spirometry measurements using clinical data and either a single-phase or multiphase lung CT.

“Rather than compute inspiratory measurements by hand, we employed the use of CNNs, which are capable of automatically determining the imaging features useful for accomplishing a given predictive task,” Hasenstab says. “For example, if a CNN believes emphysematous lung tissue is important for COPD diagnosis, which we know it is, then the CNN should automatically incorporate this information into its decision.”

In their study, the researchers emulated the PFT approach of diagnosis and staging by directly predicting PFT measurements from an inspiratory acquisition using a CNN. COPD diagnosis and staging results using the CNN-based PFT measurements were then compared with the reference standard PFT-based diagnosis and staging. “For comparison, we performed the same analysis with inspiratory and expiratory acquisitions and found that the diagnosis with a single-lung CT produces similar diagnostic and staging accuracy,” Hasenstab says. When clinical data was added, the CNN model’s predictions were even more accurate, he says.

The results showed that a CNN model developed using only a single respiratory phase CT image accurately diagnosed COPD and was also accurate within one Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage. “GOLD stage” refers to the classification system used to categorize the severity of COPD. Stages range from 1 (mild) to 4 (very severe). The lower the stage the greater the airflow obstruction.

“Our results showed accuracies of 80.7%, 79.7%, and 82.2% when using inspiratory only, expiratory only, and both inspiratory and expiratory, respectively,” Hasenstab says. “Accuracies of COPD staging within one stage were 85.8%, 84.3%, and 87.7% when using inspiratory only, expiratory only, and both inspiratory and expiratory, respectively. For both diagnosis and staging, we found that using a single CT acquisition has comparable accuracy to using both inspiratory and expiratory CT acquisitions.” The results suggest that certain markers used for COPD diagnosis may overlap across images, he says.

Hasenstab says CT scanners used for this purpose should be standard dose and have acquisition quality sufficient for formal review by a radiologist. “If the image is unfit for a radiological read, then this may negatively impact CNN accuracy,” he says. “But a formal study on image quality is needed to verify this claim.”

Hasenstab notes that a single inspiratory CT has several advantages. “Although prior studies have shown the value of expiratory CT for COPD diagnosis and staging, expiratory CT is not the clinical standard,” he says. “Implementation of an expiratory CT protocol may be infeasible at certain institutions due to the need for technologist training or other financial considerations.”

A single inspiratory CT also exposes patients to less ionizing radiation. “This is especially important for younger patients or patients requiring multiple exams over the course of their clinical care,” Hasenstab says.

A third advantage, Hasenstab says, is that the system can be applied to databases of inspiratory images that were acquired for screening purposes of other clinical indications. At-risk patients flagged for COPD can be recommended for further evaluation and a formal diagnosis.

Additional Studies Needed
There are some disadvantages to this method, Hasenstab notes. “As with any diagnostic technology, incorrect diagnoses are possible,” he says. “Moreover, the CNN was developed on the COPDGene Phase 1 cohort, which may not represent the variation of data observed for clinical care.”

Future studies are necessary to evaluate the generalizability of these methods to standard acquisition protocols, contrast-enhanced CT, and other pathologies, Hasenstab says. The group focused on COPD diagnosis and staging using only CT imaging. The research group plans further studies to investigate the synergy of using both spirometry and CT imaging for COPD prognosis, as well.

Hasenstab suspects some resistance among pulmonologists and radiologists to adopting a single-inhalation CT for staging and diagnosing COPD because it’s something different and there is often resistance to new technology. However, he says, “I am a strong advocate for developing AI-based approaches that supplement, not just replace, radiological workflow and clinical decision making.”

To the best of Hasenstab’s knowledge, the study algorithm has not been adopted elsewhere. However, his team also developed a diagnosis and staging algorithm that has been adopted at the University of California, San Diego, for research purposes pertaining to gas trapping quantification and assessment of bronchiolitis obliterans in lung transplant cases. This algorithm does require the acquisition of both inspiratory and expiratory CT. The team is also investigating ways of reducing this approach to a single acquisition. They reported on this work in Radiology: Cardiothoracic Imaging in April 2021 and Radiology Artificial Intelligence in November 2021.

So what does the future hold for CT as a diagnostic tool for COPD? It’s not likely that CT will completely replace spirometry, experts agree. It’s similar to the chest X-ray, Knapp says. “People have been predicting the death of the chest X-ray forever, and it’s 2025 and we still do chest X-rays for a number of indications,” he says. “It will be 2035 and we’ll still be using chest X-rays.”

When it comes to CT replacing spirometry, Knapp says, “I don’t think the conventional tools are going to go anywhere anytime soon. I think those will still be a part of the diagnosis and treatment. But I do think AI and CT will continue to augment low-tech spirometry and continue to make inroads.”

“Historically,” Mogel says, “CT chest imaging was seen as an adjunct to pulmonary function testing, but, more and more, now they are working together.” Mogel believes that moving forward, “Imaging will lead in certain areas, and pulmonary function testing will continue to lead in certain areas. That is the change you’re seeing happen, and that is going to continue to happen.”

— Beth W. Orenstein of Northampton, Pennsylvania, is a freelance medical writer and regular contributor to Radiology Today.