Program Information
Predicting Lung Cancer Patients` Prognosis Based On Radiomics Features From Pretreatment CT Images
L Huang*, M Fan , W Hu , J Wang , J Chen , G Qing , J Lu , Fudan University Shanghai Cancer Center, Shanghai, Shanghai
Presentations
TU-RPM-GePD-JT-3 (Tuesday, August 1, 2017) 3:45 PM - 4:15 PM Room: Joint Imaging-Therapy ePoster Theater
Purpose: The purpose of this study is to investigate the prognostic value of computed tomographic (CT) images based on radiomics features in lung cancer.
Methods: 873 radiomics features were defined, which are divided into seven groups (I: Geometry, II: Histogram, III: GLCM, IV: GLRLM, V: Wavelet GLCM, VI: Wavelet GLRLM and VII: Wavelet Histogram). For each patient, 203 radiomics features with the most representative information were analyzed. The patient database consists of two data sets: Training set (317 patients) and Validation set (54 patients). The former was downloaded from NSCLC-Radiomics collection at Cancer Imaging Archive. The latter contains patients (18 patients) treated with SBRT at Fudan University Shanghai Cancer Center ( FUSCC ) with lung cancer. We used the least absolute shrinkage and selection operator ( LASSO ) for cox regression variable selection and the 6-fold cross validation for regression model selection. Then the new-built model was validated. Concordance index (CI) was used to evaluate the performance of the radiomics model. A median threshold of predicted value was computed for each data set to split survival curve. Then we use the log-rank test to test for significant differences between two split survival curves.
Results: A radiomics model based on 3 radiomics features from Geometry, Wavelet GLCM, Wavelet GLRLM group was acquired. The training set (CI=0.628) and validation set (CI=0.581) demonstrate a fair predictive power of the radiomics model. We observed significant difference (log-rank p=0.0001) between two split survival curves in training set, demonstrating the potential association between radiomics signature and survival.
Conclusion: An effective radiomics model was acquired and validated using additional patients. Our results demonstrated the potentiality of using radiomics features for predicting prognosis and association between radiomics signature and survival. In our future work, we will further expand the database and implement the machine learning into the study.
Contact Email: