论文部分内容阅读
Purpose: Lung cancer is one of the most common forms of cancer.It is therefore highly desirable to develop methods that can give an early warning signal.The aim of this study was to explore the heterogeneous nature of textural features between malignant and benign pulmonary nodules in CT images and their correlations with pulmonary nodulesprognosis.Methods:4372 benign or malignant pulmonary nodule CT images acquired from 502 patients (152 benign cases with 1343 pulmonary nodules;350 malignant cases with 3399 pulmonary nodules) were collected.The propensity score method was used to balance observed covariates to ensure that the benign and malignant groups were balanced and comparable.By using contourlets, every pulmonary nodule can be decomposed into four scales with 48 directions, 14 texture features were thus calculated.The least absolute shrinkage and selection operator (LASSO), a type of variable selection method, was used to rank the contourlets textural features according to standardised coefficients of the LASSO regression.The first ten most important variables were evaluated for the performance of correlation with pulmonary noduleprognosis at α =0.005 based on longitudinal cases.Results: Based on 502 cases, 146 cases (73 pairs of benign and malignant cases with 1619 pulmonary nodules) were matched by their propensity score.Among the first ten most important textural features, sumentropyat sc ale 4, direction 6(Spearman correlation coefficient =-1.000, P< 0.0001) and sumentropyat scale 4, direction 15 (Spearman correlation coefficient =-0.791, P< 0.0001) were significantlydifferent for the analysis of correlation with pulmonary nodulesprognosis at at α =0.005.Conclusions: The study provided preliminary evidence that analysis of pulmonary nodules texture on CT images is potentially asuperior predictor for patients with pulmonary nodules.