A new study has found that unsupervised machine learning is able to determine radiological disease progression markers in patients with idiopathic pulmonary fibrosis (IPF). The study, published in European Radiology, also observed pattern transitions that revealed a sequence of tissue types from healthy to advanced disease.
“Besides the difficulties in diagnosing IPF, the prediction of radiological disease progression is even more challenging as disease courses in IPF are quite divergent,” the authors wrote. “It was the aim of this study to develop an unsupervised machine learning approach to identify novel radiological disease progression imaging marker patterns and evaluate if these patterns predict outcomes.”
The research team retrospectively reviewed data on 76 patients with IPF between December 2011 and October 2014 from an Italian referral center registry. They also collected a retrospective replication cohort from a center in Austria. Computed tomography (CT) scans were performed on all patients at diagnosis and at the last examination.
A machine learning model was established to analyze the imaging data and identify lung appearance patterns, markers of disease progression, and tissue transition pathways.
Read more about IPF prognosis
The results showed that the machine learning model correctly identified CT patterns associated with disease progression in IPF, and those patterns accurately predicted clinical outcomes. The model was also able to recognize a temporal sequence of scans and thereby establish a set of marker candidates that comprise a disease progression signature.
The results were replicated in the external cohort from another center, and human experts had comparable accuracy to the machine learning model in identifying radiological disease progression.
Finally, the CT data analysis revealed transition pathways from healthy lung tissue to disease, which may provide insight into the underlying mechanisms of IPF. The authors hope the data will serve as a basis for future research into disease progression and outcomes.
Pan J, Hofmanninger J, Nenning KH, et al. Unsupervised machine learning identifies predictive progression markers of IPF. Eur Radiol. Published online September 6, 2022. doi:10.1007/s00330-022-09101-x