324. FDG-PET texture analysis in NSCLC: Metabolic-pathological correlations

Abstract

Purpose in surgically resectable NSCLC prognosis is strongly affected by local invasiveness as well as unexpected lymph node diffusion found at surgery. Recently, texture analysis of PET imaging has emerged as a powerful tool to better define the metabolic behavior of neoplasms. The study is meant to evaluate the relationship of texture analysis features of surgically resected NSCLC with local invasiveness and regional sublicnical diffusion. Methods The study is a retrospective evaluation of prospectively collected data. We performed a 3D texture analysis of the PET scans on previously delineated lesion VOIs of resected stage Ia IIb NSCLC. The extraction of a list of texture analysis features was performed by an open source software package CGITA developed by Yu-Hua Dean Fang et al. [1] . We then evaluated their relationship with lymphovascular invasion, nodal upstaging and disease-free survival. Results tumours with lymphovascular invasion displayed higher levels of SUVmax, SUV entropy, normalized entropy, SUV standard deviation; and lower level of the following: normalized homogeneity, normalized inverse different moment (IDM). Tumors with nodal upstaging displayed higher levels of the following: SUV max, SUV entropy, normalized entropy, SUV standard deviation; and lower level of the following: normalized heterogeneity and normalized IDM. ROC analysis ruled out a stronger predictive power of texture analysis indicators when compared to SUVmax. Conclusions given its relationship with important prognostic indicators, texture analysis may provide a stratification of highly aggressive tumours from the very beginning of patient’s diagnostic path.