Objective:

To investigate the utility of different feature selection methods for

18

F-fluorodeoxyglucose (FDG) positron emission tomography (PET)/computed tomog

raphy (CT) radiomics features in predicting 3-year overall survival (OS) of patients with early-stage non-small cell lung cancer (NSCLC).

Methods:

From March 2017 to September 2018

the clinical and imaging data of 81 patients with clinically early-stage NSCLC

American Joint Committee on Cancer (AJCC) Ⅰ and Ⅱ stage

who underwent

18

F-FDG PET/CT imaging in The First Hospital of Shanxi Medical University and were followed up for at least 3 years were retrospectively analyzed. The radiomics features were extracted respectively from the region of interest (ROI) of PET and CT. The least absolute shrinkage and selection operator (LASSO) algorithm

mutual information (MI) algorithm

recursive feature elimination (RFE) algorithm and univariate Cox regression analysis are used to select features for constructing Cox proportional risk models. After internal validation

the consistency index (C-index) was used to evaluate the efficiencies of each model in predicting the 3-year OS in NSCLC patients.

Results:

Radiomic features were strongly correlated with survival outcomes and could be used as analysis variables; compared with other methods

the Cox proportional risk models constructed by LASSO has the highest C-index (0.830.07).

Conclusion:

The feature selection method directly affects the efficiencies of model

and a suitable method should be found to improve the efficiencies when using radiomic features.