Temporal Stability and Prognostic Power of Radiomic Features for Survival Analysis in Lung Adenocarcinoma

The main aim of our study is to estimate the temporal stability and assess the prognostic performance of radiomic features extracted from lung computer tomography (CT) images. We have considered the Lung CT diagnosis dataset for our work which contains 61 patients identified with lung adenocarcinoma cases. In this work, we have segmented 284 nodules by applying the random walk ensemble segmentation technique, and thirty-eight radiomic features were extracted from the segmented nodule region. These features include 19 gray-level co-occurrence (GLCM)-based features and 7 gray-level run length matrix (GLRLM)-based features, 12 histogram-based features. Later, the temporal stability was explored by considering Intra-class correlation coefficients (ICC) between features extracted from segmented nodule regions using our proposed segmentation technique and the segmented ground truth images provided by radiologists in the LungCT-Diagnosis dataset publicly available in The Cancer Imaging Archive (TCIA). The prognosis performance of features with temporal stability was assessed based on the Kaplan-Meier survival analysis. It has been observed that 16 radiomic features exhibited temporal stability, and seven temporally stable features have a statistically strong prognostic association with patient survival. This work explores the temporal stability and prognostic power of radiomic features using survival analysis to achieve optimum treatment planning at an early stage of diagnosis in lung adenocarcinoma cases.