AUTHOR=Zhang Tao , Zhang YueHua , Liu Xinglong , Xu Hanyue , Chen Chaoyue , Zhou Xuan , Liu Yichun , Ma Xuelei 

TITLE=Application of Radiomics Analysis Based on CT Combined With Machine Learning in Diagnostic of Pancreatic Neuroendocrine Tumors Patient’s Pathological Grades

JOURNAL=Frontiers in Oncology

VOLUME=Volume 10 - 2020

YEAR=2021

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2020.521831

DOI=10.3389/fonc.2020.521831

ISSN=2234-943X

ABSTRACT=Purpose：The purpose of this study was to explore the correlation between CT texture features and the pathological grading of Pancreatic Neuroendocrine tumors.
Materials and methods: A retrospective study was conducted on 54 patients (32 males and 22 females; average age (55.3 +11.1 years) with Pancreatic Neuroendocrine tumors (from March 2011 to May 2018) All patients had definite pathological diagnosis and grading results. The texture parameters of tumors were extracted from enhanced CT images taken before treatment. All parameters were analyzed by ROC curve, and the parameters with the largest AUC and P < 0.05 under each texture parameter category were selected. The selected parameters are modeled by binary logistic regression, and three models are constructed: image model, texture model and combined model. ROC was used to test the diagnostic capacity of these models.
Result：We get the most discriminating image-based parameter (maxvalue, AUC: 0.678, p=0.037) and 5 most discriminating texture parameters (SHAPE_Volume (# vx), GLCM_Correlation, GLRLM_RLNU, NGLDM_Coarseness and GLZLM_ZLNU, AUC: 0.703,0.706, 0.715, 0.676, 0.702, p< 0.05) were selected by ROC analysis. Among them, maxvalue, SHAPE_Volume (# vx), GLCM_Correlation, GLRLM_RLNU and GLZLM_ZLNU tends to indicate pathological high-grade. NGLDM_Coarseness tends to indicate pathological low-grade. The AUC of image-based model is 0.678, with sensitivity of 81.1%, specificity of 52.9%. The AUC of texture model is 0.749, with sensitivity of 51.4%, specificity of 94.1%. The AUC of combined model is 0.766, with sensitivity of 56.8%, specificity of 94.1%. 
Conclusion: The preoperative enhanced CT image texture analysis to predict the pathological grade of pancreatic neuroendocrine tumors has a potential application.