Research And Implementation Of Deformation-based Feature Extraction In Medical Images

Brain tumor is the most serious disease in the department of neurosurgery,it will squeeze or destroy other normal brain tissues,thereby endangering the lives of patients.Medical image processing technology can help the computer to automatic segment brain tumor,which greatly reduces the time required for manual and the subjective deviation.Therefore,it is great significance to use computer technology to help doctors research and segment brain tumor.But in the current brain tumor automatic segmentation system,the input features such as grayscale value or edges,they can not well reflect the position of brain tumor,so the accuracy of these automatic segmentation systems are not up to the requirements of the doctor is looking for,which is the biggest bottleneck of current brain tumor automatic segmentation system.Aiming at these problems,this paper propose a deformation-based feature extraction algorithm,this algorithm can extracts the deformation-based feature which can better reflect the position of brain tumor.The main work and contributions of this paper are as follows:First,the relationship between brain tumor and normal tissues is researched.It can find that brain tumor will squeeze normal tissues and normal tissues will produce a corresponding deformation and displacement in the opposite direction.So we can use the deformation information of normal tissues to indirectly reflect the position of brain tumor.On this basis,a deformation-based feature extraction algorithm is proposed in this paper.This algorithm can extracts quantitative deformation information of normal tissues and change it into deformation-based feature which can be used for brain tumor segmentation system.This algorithm mainly includes three steps,extracting intracranial tissue shape,obtaining 3-D control points,deformation modeling and feature extraction.The main task of these steps are extracting intracranial tissue shape from nuclear magnetic resonance images,obtaining 3-D control points from template and deformed tissues,using 3-D control points to deformation modeling and feature extraction.Finally,this paper use artificial neural network and support vector machine as supervised pattern recognition method and fuzzy C-means clustering as unsupervised pattern recognition method to verify the validity of the deformation-based feature.In all of the experimental method,through the quantitative comparison between include and not include the deformation-based feature,it can be seen that all statistical indicators have correspondingly improved after adding the deformation-based feature,the results show that the deformation-based feature can really help brain tumor segmentation system.In this paper,a new research direction of feature extraction in medical image processing is proposed,which provides a useful exploration and reference for the medical image segmentation and human tissue shape change analysis.