Researchers have developed a fully automatic, deep learning-based model that is able to predict survival in idiopathic pulmonary fibrosis (IPF) by estimating total lung capacity (TLC) from chest radiographs.

Higher estimated TLC percentage at baseline was associated with a lower risk of mortality in patients with IPF, with an adjusted hazard ratio of estimated TLC percentage of 0.97%.

The prognostic discrimination performance was maintained after substituting estimated TLC for forced vital capacity.

Continue Reading

“It is noteworthy that our model was able to estimate TLC accurately and reproducibly with use of a single chest radiograph,” the researchers said.

Read more about IPF prognosis

The mean absolute difference and within-subject standard deviation between observed and estimated TLC ranged from 0.52 to 0.69 L and 0.53 to 0.73 L, respectively. The repeatability coefficient between the observed and estimated TLC ranged from 1.46 to 2.03 L. Moreover, in test-retest agreement, the repeatability coefficients for 2 consecutive chest radiographs ranged from 0.64 to 0.67 L.

According to the authors, using chest radiographs to estimate TLC has advantages, including getting rapid results at no additional cost or need for additional human resources.

The researchers used different data sets to develop and validate the method. The pretraining data set included 50,000 computerized tomography scans. The development data set to fine-tune the method included 3523 pairs of same-day TLC and posteroanterior chest radiographs from 2483 patients. The researchers used 2 external test sets comprising 207 and 216 patients. Test-retest agreement was analyzed in 65 and 88 patients from each external test set.

To evaluate the prognostic value of the estimated TLC, the researchers used data from 217 patients with IPF.


Kim H, Jin KN, Yoo S-J, et al. Deep learning for estimating lung capacity on chest radiographs predicts survival in idiopathic pulmonary fibrosis. Radiology. Published online October 25, 2022. doi:10.1148/radiol.220292