The Research Of Diagnosis Model For Gliomas' Grade Using MR Spectroscopy

This study aimed to build a model for medical assistant diagnosis and carry out auto-grading in glioma, using statistical analysis and Fuzzy maths theory. All what we want was to provide clinicians with tools for glioma grade diagnosis by MRS information. The whole paper includes three parts as follows.In the first part, we used MeSH search in PUBMED and used Key words search in CNKI databases to indentify related MRS information for glioma grade and selected 14 randomized controlled studies as the objects of Meta-analysis. The specificity and sensitivity of MRS variables in glioma grade was obtained by Meta-analysis. Also we compared relevant parameters to gain evidence-based medical message for selecting characteristic variables for fuzzy clustering analysis and model building.In the second part, we drew characteristic variables of MRS for fuzzy clustering analysis to obtain a clustering result of glioma in different grades. In this process, original data were standardized first and similar matrix was carried out by calculating cosine of angle or maximum and minimum distance. Then similar matrix was transformed into equivalence matrix by using Washall algorithm. Through N cycles of clustering, the classification of glioma was completed and all levels of fuzzy subsets were obtained. Finally, based on the fuzzy subsets, we analyzed the unclassified elements and realized their automatic classification.In the third part, based on the fuzzy clustering algorithm, we built an assistant diagnostic model for brain tumor through automatic extraction of membership function and IF-then diagnosis rules drew later. Then we can obtain the relationship between input and output process in glioma diagnostic classification.In short, we can build a brain tumor classification model through extracting characteristic variables of MRS and fuzzy clustering algorithm. And the experimental results show that the model can classify glioma accurately, with the practical value of clinical application. This may provide methods for further researches on more complex brain tumor diagnosis models.