SU‐F‐R‐45: The Prognostic Value of Radiotherapy Based On the Changes of Texture Features Between Pre‐Treatment and Post‐Treatment FDG PET Image for NSCLC Patients

Abstract

Purpose:The purpose of this research is investigating which texture features extracted from FDG‐PET images by gray‐level co‐occurrence matrix(GLCM) have a higher prognostic value than the other texture features.Methods:21 non‐small cell lung cancer(NSCLC) patients were approved in the study. Patients underwent 18F‐FDG PET/CT scans with both pre‐treatment and post‐treatment. Firstly, the tumors were extracted by our house developed software. Secondly, the clinical features including the maximum SUV and tumor volume were extracted by MIM vista software, and texture features including angular second moment, contrast, inverse different moment, entropy and correlation were extracted using MATLAB.The differences can be calculated by using post‐treatment features to subtract pre‐treatment features. Finally, the SPSS software was used to get the Pearson correlation coefficients and Spearman rank correlation coefficients between the change ratios of texture features and change ratios of clinical features.Results:The Pearson and Spearman rank correlation coefficient between contrast and SUV maximum is 0.785 and 0.709. The P and S value between inverse difference moment and tumor volume is 0.953 and 0.942.Conclusion:This preliminary study showed that the relationships between different texture features and the same clinical feature are different. Finding the prognostic value of contrast and inverse difference moment were higher than the other three textures extracted by GLCM.