The Enhanced Value of Parameter Response Mapping in a One-Stop CT Scanning: Predicting Pulmonary Function Test with Machine Learning

Abstract

Background: To evaluate the performance of a machine learning approach to predict pulmonary function test (PFT) result from parameter response mapping (PRM), which enhances the value of one-stop CT scanning. Methods: 615 subjects with PFT and paired inspiratory and expiratory chest CT scanning were enrolled retrospectively, and classified into normal group, high risk group and COPD group based on PFT. 72 PRM-derived quantitative parameters including volume (cc) and volume percentage (%) of emphysema, functional-small airways disease (fSAD), and normal lung tissue (Normal) were acquired. Random forest regression model was constructed and tested on the aspect of PFT prediction, and the classification performance based on these PFT predictions was further evaluated. Findings: The machine-learning model based on PRM parameters showed good performance to predict PFT, with a coefficient of determination (R2) of 0·749 and 0·792 for FEV1/FVC and FEV1% regression respectively. The sensitivity, specificity, and accuracy for differentiating normal group from high-risk group were 0·85, 0·90 and 0·88, respectively; and 0·89, 1 and 0·99, respectively, for differentiating non-COPD group from COPD group. Interpretation: A machine learning model based on CT-derived PRM parameters could enhance the clinical value of one-stop chest CT scanning by predicting PFT results. The technique has the potential to benefit patients in clinical practice, especially for those with difficulty to perform PFT. Funding Statement: This work was supported by the National Natural Science Foundation of China [grant number 81871321, 81930049]; the National Key R&D Program of China [grant number 2016YFE0103000, 2017YFC1308703]. Declaration of Interests: None. Ethics Approval Statement: This study was approved by the hospital ethics committee and the informed consent of patients was waived.