Data Analysis Of Complex Tumor Based On DCE-MRI

In clinical diagnosis,dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)can provide blood supply and different protions of lesions after tracer injection which can not be reflected by conventional magnetic resonance imaging,therefore DCE-MRI has gradually become part of routine MRI examination in most hospitals.The time-intensity-curves(TIC)of DCE-MRI can reflect hemodynamic features of lesion,if we can cluster the region of interest(ROI)on the basis of time-intensity-curves,then we will get different parts of lesion which can be analyzed as benign or malignant through some criterion,and this will provide aid for doctor diagnosis.This dissertation starts with the aim of exact segmentation of regions of interest,an improved variational level set method which can handle intensity inhomogeneity of medical images is introduced for this,on the basis of this method,a scale stopping function was added to modify,this function is not only able to speed up evolution rate of curves,but also can effectively prevent leakage boundary.Some experiments have been done to prove the promotion of segmentation accuracy,computation period and iteration of the improved level set.To optimize the result of level set,signal intensity ratio(SIR)and initial signal intensity and seed region growing method are introduced to assist improved level set for getting some more accurate segementation results.After getting the region of interest,in order to distinguish benign and malignant,some kinds of clustering algorithms are introduced to distinguish benign and malignant regions of lesion,including K-means method,Fuzzy C-means method and convex analysis of mixture(CAM)method.Using these three methods,experiments have been done in different cases including breast cancer and prostate cancer from different hospitals.Under the assumption of two or three clustering centers,we can also get clustering center curves and pcolor spatial distribution maps.After analyzing the dispersity of clustering centers and the richness of spatial distribution information,it shows that CAM method is able to get much distinguishing clustering centers which can represent different component of lesion and pcolor spatial distribution maps of Fuzzy C-means method and CAM method contains richer distribution information which can reflect different component distribution of lesion in space.The results above have been confirmed by radiologists,and the three methods all have certain auxiliary significance in clinical imaging diagnosis.