Research On Lesion Segmentation Of Various CEST Magnetic Resonance Images In A Rat Model

Chemical exchange saturation transfer(CEST)is a new magnetic resonance imaging technology,which is very sensitive to chemical exchange of metabolites.In this technique,a saturation pulse "label" is applied to the exchangeable protons in a specific molecule.Then the saturation signal is transferred to water through chemical exchange and amplified.Finally,the molecules are indirectly detected by the change of water signal.However,due to the low SNR of CEST saturated image and the prone to artifacts,the injured area can not only be extracted by signal threshold.However,manual sketching is a time-consuming and error prone process for the extraction of massive image data.In order to meet this demand,this paper uses different pixel clustering methods to automatically extract the injured areas of various kinds of CEST images,and initially explores the automatic detection method of brain stroke rat model based on the clustering results.In this paper,based on the 24-hour and 6-hour high field CEST MRI data of rats with cerebral apoplexy after B0 field correction and image registration,S0 image,MTRasym and Z-spectrum original saturated image are selected respectively,and different pixel clustering methods are used to realize the automatic extraction of brain injury area.The main contents of the research are as follows:(1)Based on K-means pixel clustering,the brain injury area segmentation method is used to segment and extract the brain injury area of 24-hour rat image.In the process of clustering,the optimal number of clusters is calculated by G-value optimization method.Finally,two-dimensional correlation coefficient was used to quantify the similarity between each ppm image,and the segmentation results with gold standard.(2)Based on the Fuzzy C-Means clustering method,the early brain injury image of rats was extracted,and the lesions were segmented on the 6-hour images of rats with no obvious injured area.In this paper,we not only compared and calculated the similarity of 6-hour and24-hour rat image segmentation results,and proved that FCM algorithm can achieve effective lesions extraction for 6-hour rat image.The paper also compares the similarity between FCM and K-means to show the superiority of FCM.(3)Prediction and classification of early brain injury by CEST based on support vector machine.By NSCT and SVM,the brain images of 6-hour normal or abnormal rats were automatically predicted and classified on the basis of 24-hour image data.By comparing the classification performance of different kernel function SVM,it is found that Gaussian radial basis kernel SVM can predict the early CEST data more accurately.Through a series of experimental analysis,this paper fully proves that the proposed methods have achieved good results in the segmentation and prediction of rats' CEST MRI images,and provide a theoretical basis for the future work of clinical detection.