Assessment of Prognosis of Patients With Idiopathic Pulmonary Fibrosis by Computer-aided Analysis of CT Images

Purpose: To present the Gaussian Histogram Normalized Correlation (GHNC) system, to quantify the extent of disease on computed tomography (CT) images of idiopathic pulmonary fibrosis (IPF), and to assess its utility by comparing the radiologist' scoring and prognosis. Materials and Methods: GHNC was used to analyze baseline thin-section CT images (30-60images per patients) of 40 patients with IPF. It classified the CT lung field into normal (N), ground-glass opacities (G), consolidation (C), emphysema (E) and fibrosis (F) patterns [the latter was also subdivided into reticular (F1) and honeycomb (F2) patterns], then the relative lung volume and relative each pattern volume (area multiplied by slice thickness and interval and divided by predicted total lung capacity) were estimated. The radiologists estimated the area of these patterns on 4 slices per patient; the average was regarded as total extent of each pattern. We compared the estimates determined by radiologists and GHNC with pulmonary function tests, and used Cox regression analysis to examine the relationships between the Volumes and patient survival. Results: The area of each pattern measured by GHNC correlated significantly with that estimated by the radiologist on 160 images (P < 0.001, each). The volumes of N-pattern and F-patterns are measured by GHNC correlated with carbon monoxide diffusing capacity. During the follow-up (mean 49mo), 24 patients died. Relative lung volume, N-pattern, F-pattern and F2-pattern correlated with survival in univariate analysis. Multivariate analysis identified F2-pattern volume by GHNC (P = 0.034) as a significant predictor of survival. Conclusions: The GHNC provides automatic measurement of volume of fibrosis. The F2-pattern on CT can predict prognosis of patients with IPF.