Deep Learning for Estimating Lung Capacity on Chest Radiographs Predicts Survival in Idiopathic Pulmonary Fibrosis

• DOI: 10.1148/radiol.220292
• Publication Date: 2022-10-25T21:16:45Z
• Authors: Hyungjin Kim; Kwang Nam Jin; Seung-Jin Yoo; Chang Hoon Lee; Sang-Min Lee; Hyunsook Hong; Joseph Nathanael ...; Soon Ho Yoon
• Keywords: Male; Middle Aged; Idiopathic Pulmonary Fibrosis; 3. Good health; Radiography; 03 medical and health sciences; Deep Learning; 0302 clinical medicine; Humans; Lung Volume Measurements; Lung; Aged; Retrospective Studies

Background Total lung capacity (TLC) has been estimated with use of chest radiographs based on time-consuming methods, such as planimetric techniques and manual measurements. Purpose To develop a deep learning–based, multidimensional model capable estimating TLC from demographic variables validate its technical performance clinical utility multicenter retrospective data sets. Materials Methods A learning was pretrained 50 000 consecutive CT scans performed between January 2015 June 2017. The fine-tuned 3523 pairs posteroanterior plethysmographic measurements patients who underwent pulmonary function testing the same day. tested sets two tertiary care centers one community hospital, including (a) an external test set 1 (n = 207) 2 216) for (b) idiopathic fibrosis 217) utility. Technical evaluated various agreement measures, assessed in terms prognostic value overall survival multivariable Cox regression. Results mean absolute difference within-subject SD observed were 0.69 L 0.73 L, respectively, (161 men; median age, 70 years [IQR: 61–76 years]) 0.52 0.53 (113 63 51–70 years]). In (145 67 61–73 years]), greater percentage associated lower mortality risk (adjusted hazard ratio, 0.97 per percent; 95% CI: 0.95, 0.98; P < .001). Conclusion fully automatic, learning–based total radiographs, predicted fibrosis. © RSNA, 2022 Online supplemental material is available this article. See also editorial by Sorkness issue.