Improving functional correlation of quantification of interstitial lung disease by reducing the vendor difference of CT using generative adversarial network (GAN) style conversion

This study evaluated whether applying a generative adversarial network (GAN)-based style conversion (RouteGAN) could reduce inter-vendor differences in interstitial lung disease (ILD) quantification, thereby improving the functional correlation of quantitative CT (QCT) measurements. Researchers analyzed 112 idiopathic pulmonary fibrosis (IPF) patients to compare ILD quantification accuracy before and after CT style conversion. The transformed CT images showed stronger correlations between reticulation, fibrosis scores, and pulmonary function test (PFT) indices than the original CT images. Radiologists also rated the quantification accuracy higher for the transformed CTs than for the original ones. Aview Lung Texture software was used for ILD quantification, contributing to more precise ILD pattern analysis on style-transformed CT images. The study suggests that CT style conversion can enhance the consistency of ILD quantification across different CT vendors.