PUB101 High Predictive Value of Invasiveness for IPAs from Preinvasive Lesions Appearing as Pure Pround-Glass Nodules on Ultra-High Resolution CT

Abstract

To analyze the correlations between histological invasiveness and radiologic findings of early invasive pulmonary adenocarcinomas (IPAs) and preinvasive lesions appearing as pure ground-glass nodules on Ultra-HRCT. To evaluate the potential predictive factors of invasiveness for pure ground-glass nodules(PGGN)on Ultra-HRCT. Retrospective analysis of 123 lesions (16 were Atypical adenomatous hyperplasia (AAH), 35 were Adenocarcinoma in situ (AIS), 35 were minmally invasive adenocarcinoma (MIA), 37 were Invasive adenocarcinoma (IA) with PGGN on Ultra-HRCT from January 2014 to June 2014 in a single-central hospital. Only one lesion can be enrolled in every patient. There were 93 females and 30 males, with a median age of 58 (24-77) years old. All focuses were resected and confirmed by pathology. The clinical data (gender, age) and Ultra-HRCT findings(lobulation, spiculation, pleural indentation, aterial gathering, bubbles/air bronchogram, shape, margin, internal uniformity and tumor-lung interface, size of lesion, average density of lesion, the corresponding lung’s average background density)were recorded, pathological data of the 123 cases according to the new lung adenocarcinoma classification proposed by the IASLC/ATS/ERS were recorded. With respect to the correlation between histological invasiveness and the morphological features of PGGN, only tumor-lung interface (P>0.05) had no correlation with histological invasiveness, and its predictability was quite low (Lambda value, λ=0). The rest of the morphological features, including lobulation, spiculation, pleural indentation, aterial gathering, bubbles/air bronchogram, shape, margin, and internal uniformity were specific factors that differentiated invasive adenocarcinoma from AIS or AAH (P<0.05). The speculation and internal uniformity showed a good correlation and predictability with the Lambda values (cross validated correlation coefficient) of 0.519 and 0.568, the odds risk of spiculation, bubbles/air bronchogram and internal uniformity were 36.36, 28.37 and 22.5, respectively. In ROC curve analysis, the area under curve (AUC) of maximum lesion area on CT scan, lesion size in cranial-caudal direction, the relative average density and the actual average density were 0.841, 0.827, 0.74, and 0.734, respectively. The optimal cut-off values of these four factors were 56.5 mm2, 10.5mm, 276HU and -539Hu. Binary Logistic Regression procedure offered two kinds of stepwise methods for selection of the “best” predictors to include in the mode, and these predictors were speculation, internal uniformity, lesion size in cranial-caudal direction, the actual average density of lesion and gender. Then constructed an ROC Curve according to the regression model, and the higher AUC(0.95) reveal excellent differentiating accuracy of above predictors. The Ultra-HRCT morphological features of PGGN were significant correlated with histological invasiveness of IPAs and preinvasive lesions except for tumor-lung interface. IPAs can be accurately differentiated from preinvasive lesions by using Ultra-HRCT.