Brain Magnetic Resonane Image Tumor Segmentation With The Use Of Intracranial Tissue Deformation

Intracranial tumor is a typical serious nervous system disease to human health.Although automated segmentation is of significant importance by alleviating the manual work and reducing variability associated with subjective human judgment.However accurate automated intracranial tumor segmentation is a classic challenging task and desired segmentation accuracy is still out of reach.In the brain tumor segmentation system,extracting features which represent the properties of tumor and applying them as part of input data for segmentation algorithm is of great importance for the accuracy of brain tumor segmentation.However one disadvantage of the features in the traditional brain tumor segmentation systems is that,the normal features such as grayscale value and textures they used do not contain strong correlation with brain tumor.According to study,intracranial lateral ventricles(LaVs)can be deformed by the compression from brain tumors,correct extraction and application can enhance brain tumor segmentation accuracy.Based on these study and ideas,by addressing related research problems,this thesis proposes a new method for extracting features of 3-dimensioanl intracranial lateral ventricular(LaV)deformation,and utilizing anatomical meaning between intracranial tissues in a higher level,thereby further improves brain tumor segmentation accuracy.In this thesis,several important problems in the magnetic resonance(MR)image,brain tumor and LaVs are reviewed.Based on the observations that brain LaVs are compressed by tumors,this research investigates the correlation between LaV deformation and tumor location.It focuses on the design and implementation of a feature extraction component which transforms 3-dimensioanl LaV deformation into an additional feature for improving accuracy of brain tumor segmentation.This component is mainly composed of severl processes of LaV shape retrieval,3-dimensional LaV alignment,LaV deformation quantification and feature data transformation.To ensure accurate and reliable LaV deformation data,we proposed an improved dynamic wavelet fuzzy-c means(dwFCM)algorithm and a method for creating 3-dimensional template LaVs.Respective experiments demonstrate the validity of the two methods.From the observation of extracted LaV deformation feature,this feature has a strong correlation with brain tumor.By using the classifiers of pattern recognition algorithms which support multi-dimensional dataset,we combine the extracted LaV deformation feature with other low-level features such as MR image graylevel and texture,and apply the dataset to general brain tumor segmentation system.Quantitative analysis on the multiple comparative experiments suggests that,the extracted LaV deformation feature is useful in enhancing brain tumor segmentation accuracy at various levels.This result further suggests the correlation between LaV deformation and the existence of brain tumor.Comparing with traiditional brain tumor segmentation systems,the major advantage of this system is the LaV deformation feature represents the anatomical meaning of intracranial tissue.And the feature extracted in the 3-dimensional way is more reasonable compared to the existing work of that from 2-dimensional images.This research not only provides theoretical basis for the study of intracranial tissue or organ shape,it is also helpful in enhancing the level of medical areas of tissue shape analysis,image guideded surgery and tumor growth evaluation,and computer science areas of pattern recognition and artificial intelligence.