Radiomic Features Selection From PET/CT Images for the Adenocarcinoma Histologic Subtype Identification in Non-small Cell Lung Cancer

The aim of the present study is to contribute to medical diagnostic by applying a selection method of relevant Radiomic features from PET/CT images to help identifying adenocarcinoma in Non-Small Cell Lung Cancer (NSCLC). This work is based on radiomics techniques that allows qualitative and quantitative high performance analysis, from the calculation of radiological and molecular characteristics from PET/CT images. The Chang-Gung Image Texture Analysis (CGITA) was used to extract texture features based on morphological characteristics (shape, volume, surface area, density and mass), statistics (attenuation histogram) and regional (intra-tumor neighborhood analysis) in the region of interest (ROI) using the maximum, mean and metabolic volume standardized uptake values (SUV) considering tumor volume semi-automatically segmented ally. The CGITA returned 72 features of 24 selected images from The Cancer Imaging Archive TCIA database. They were analyzed using principal component analysis (PCA) to reduce data dimension in order to optimize computational effort and make adenocarcinoma identification more efficient. The results showed that three features based on co-occurrence matrix (contrast, entropy and dissimilarity) were responsible for more than 95% of the full variance of the data.