Study on Prognosis Factors of Non-Small Cell Lung Cancer Based on CT Image Features

This study aims to investigate the prognosis factors of non-small cell lung cancer (NSCLC) based on CT image features and develop a new quantitative image feature prognosis approach using CT images. Firstly, lung tumors were segmented and images features were extracted. Secondly, the Kaplan-Meier method was used to have a univariate survival analysis. A multiple survival analysis was carried out with the method of COX regression model. Thirdly, SMOTE algorithm was took to make the feature data balanced. Finally, classifiers based on WEKA were established to test the prognosis ability of independent prognosis factors. Univariate analysis results reflected that six features had significant influence on patients' prognosis. After multivariate analysis, angular second moment, srhge and volume were significantly related to the survival situation of NSCLC patients ( P < 0.05). According to the results of classifiers, these three features could make a well prognosis on the NSCLC. The best classification accuracy was 78.4%. The results of our study suggested that angular second moment, srhge and volume were high potential independent prognosis factors of NSCLC.