Machine learning algorithms could predict onset of 'popcorn lung' post-transplant

A combination of machine learning and quantitative computed tomography (CT) could predict the eventual onset of bronchiolitis obliterans syndrome (BOS), also known as popcorn lung, in patients receiving lung transplants, researchers  reported in Academic Radiology this week.

“Lung transplantation is an important treatment option for patients with advanced, irreversible lung disease,” corresponding author Eduardo J. Mortani Barbosa Jr., MD, and colleagues wrote. “However, despite continued advancements in lung transplantation techniques, long-term survival is limited by chronic lung allograft dysfunction.”

This dysfunction comes in the form of BOS, they said, which is a commonality in the world of lung disease. Patients see only a moderate chance of survival for 2.5 years after a lung transplant, and that risk continues indefinitely, with a cumulative incidence of more than 50 percent after 5 years.

Despite the urgency of the complication, clinicians haven’t been able to find an effective treatment for BOS, Mortani Barbosa et al. wrote. While surgical lung biopsies were the standard in the past and evidence supporting augmented immunosuppression has shown promise for the future, a BOS cure doesn’t exist.

The authors wrote that imaging isn’t an established diagnostic tool for evaluating BOS, but research has suggested standard, high-resolution CT (HRCT) scans could be a helpful tool for managing post-transplant patients. So, for their own work, Mortani Barbosa and his co-authors combined quantitative CT metrics and functional respiratory imaging (FRI), which is a method based on HRCT quantitative computational image processing of lung parenchyma and airways.

The team tracked post-transplantation changes in lung and airway structure and function in a population pool of 71 patients. All participants had undergone a lung transplant, according to the study, and more than half had developed BOS—41 patients versus the BOS-free subgroup, which was 30 patients strong.

The BOS cohort saw a decline in forced expiratory volume in the first second (FEV) of greater than 10 percent when compared to baseline FEV, the researchers reported. That reduction correlated with an increase in lung volume and in the central airway volume at function residual capacity, which wasn’t the case in the non-BOS population, who saw a decrease in central airway volume at total lung capacity with declining FEV.

The authors found that 23 baseline quantitative CT parameters could distinguish between non-BOS patients and those who eventually developed the condition—better than any pulmonary function testing parameters have ever been able to. Adding the use of machine learning meant the researchers could identify BOS developers at baseline with an accuracy of 85 percent and using just three quantitative CT parameters.

“This approach may become useful to optimize management of lung transplantation patients,” Mortani Barbosa and colleagues wrote. “By using quantitative CT methods such as FRI, changes in the transplanted and nontransplanted lungs can be assessed separately to improve prediction of early BOS onset.”