[Radiological evaluation of advanced gastric cancer: from image to big data radiomics]

Following the increased demand of personalized medicine to precise radiology in advanced gastric cancer, there is particular need for objective and powerful surrogate to help the gastro-radiology to break through the bottleneck of imaging resolution and the defect of subjective diagnosis, which can further improve the efficacy of staging and response evaluation. On the basis of the existing imaging resolution, the radiomics can perform massive data mining through texture analysis and big data, using artificial intelligence deep learning and other algorithms to screen and integrate images and clinical features for modeling and diagnosis, which may further improve the efficacy of staging and response evaluation theoretically. In this paper, we focused on gastro-radiology and radiomics, and reviewed five dimensions progressively: (1) As the first choice for staging and response evaluation, CT application is limited by radiologists' ability to excavate image features and information integration, which needs more powerful image processing method. (2) Radiomics texture analysis can provide massive objective image information that can not be identified by the radiologists' naked eye. It is more detailed and provides quantitative evaluation of the characteristics of tumors better than the radiologists' subjective vision analysis, which can dig potential microscopic information. In the recent two years, the research on the application has been progressing rapidly, covering almost all the solid tumors, and solving the various clinical focuses using entropy, skewness, heterogeneity and other texture analysis indicators. (3) The research progress of radiomics in gastric cancer from the following three directions was summarized: differential diagnosis and biological behavior analysis, staging, and response prediction and evaluation. The current research confirmed the high efficiency of radiomics and texture analysis in differentiating different types, stages and responders of gastric cancer, which can act at least as an important supplement for the subjective evaluation of the radiologists.(4) The congenital defects of radiomics and the current problems on research were summarized, in order to avoid misuse and pitfalls. (5) The radiologists need not to worry about being replaced in the expectation of the future AI radiomics; on the contrary, AI radiomics will be a good assistant. The radiologist should actively take part in the MDT and cooperate with multi-center colleagues to promote the development of large data radiomics in gastric cancer.