Scientists to Compete at Algorithms for Predicting Lung Decline in IPF
The Open Source Imaging Consortium (OSIC), a nonprofit collaborative group focused on combatting lung diseases, has launched a competition aiming to create artificial intelligence (AI) programs that can help to predict lung function decline in people with pulmonary fibrosis.
Called the OSIC Pulmonary Fibrosis Progression Challenge, it is being administered by the data science community platform Kaggle, and is open to all researchers and clinicians wishing to participate either individually or in teams.
“OSIC was created to bring divergent groups together to look at new ways of fighting complex lung disease,” Elizabeth Estes, the executive director of OSIC, said in a press release. “In addition to utilizing expertise from academia, industry and philanthropy, we wanted to introduce clinicians to the broader artificial intelligence and machine learning community to see if new eyes and new tools could help us move forward, faster. We are excited to see the progress that can be made for patients all over the world.”
According to its website, OSIC’s “mission” is to unite radiologists, clinicians, and computational scientists worldwide in working to improve imaging-based treatments.
Idiopathic pulmonary fibrosis (IPF) is a disease of unknown cause characterized by progressive fibrosis (scarring) of the lungs, which impairs lung function. This scarring is visible on CT scans of the chest.
One challenge of caring for people with IPF is the difficulty in predicting how the disease will progress. That is, it’s hard to estimate how quickly an individual’s lung function is likely to decline.
“The heterogeneity of outcome in this disease complicates clinical decision making for individual patients, increasing their anxiety and fear,” said Kevin Brown, MD, a pulmonologist (lung doctor) at National Jewish Health who serves as OSIC’s lead pulmonologist.
Broadly, challenge participants will be tasked with creating AI programs that predict changes in lung function over time for people with IPF.
To build and train their algorithms, they will be given lung CT scans, taken at diagnosis, from IPF patients. Participants will also receive clinical information on the patients, as well as forced vital capacity (FVC) — a standard measure of lung health — collected over about two years of follow-up.
These scientists will then test their programs on a second group of patients, for whom they have access only to an initial lung CT scan, clinical data, and FVC measurement. Their goal is use their AI program to generate predictions about changes in lung function for these people — namely, what their final three FVC measurements will be.
Competitors’ predictions will be checked against real-world data from the test group, and specialized statistical algorithms will determine which predictor is best.
“Success in this challenge will help clinicians provide clarity to our patients, and ultimately improve treatment, trial design and accelerate the clinical development of novel therapies,” Brown said.