Research On Brain CT Image Classification Method And Visualization

With the rapid development of science and technology,as well as the continuous popularization of digital medical,large numbers of medical data are produced every day.Effective analysis and mining of these data can contribute to the diseases prevention and detection so as to help doctors for treatment.In the study of brain diseases,the classification of brain CT images based on historical data can conduct early prevention and treatment of some lesions.It can not merely improve the cure rate,but assist doctors in the diagnosis of brain diseases.In the brain CT image generation process,due to machine performance or scanning technology and other reasons,the image quality are of differentiation.Thus the preprocessing of CT images is an essential stage before data analysis.Moreover,with the increase of data volume,the artificial preprocessing is hard to meet the demand.For example,the medical image with symmetrical structure,such as brain scanning image,the determination of its symmetry axis is especially important in preprocessing.Therefore,for large-scale preprocessing of brain CT images,the priority is to extract the brain ideal mid-sagittal line(i ML).According to the i ML,the migration of the original image can be calibrated so as to improve the accuracy of classification.Classification is an important part of data mining and machine learning.It is widely used in meteorological analysis,financial analysis and medical data analysis.In the research of brain CT image classification some methods are commonly used,including statistical learning methods,neural networks,pattern recognition and machine learning methods.The existing methods can obtain good classification results in some specific fields based on sufficient training samples.However,in practical applications,the problem of uneven sample distribution and insufficient training samples is often encountered.These methods are difficult to get satisfactory classification accuracy under the lack of sufficient training.As an important way of data analysis,visual analysis has been widely used in computer-aided diagnosis and medical information management in medical data analysis.However,the visualization system based on large scale data analysis is mainly visualization of the original data and results,lack of certain interaction,for users is still a "black box".To increase the participation of users in the system,can help the user to facilitate the understanding of the algorithm,the discovery and exploitation of new knowledge in complex data.Besides,for ordinary users,it can be more intuitive to observe the whole process of clinical diagnosis.Therefore,according to the above problems,the main work of this paper is as follows:(1)In this paper,a method of ideal mid-sagittal line(i ML)extraction method based on scale invariant feature transform(SIFT)features is proposed.Our proposed method is independent from the symmetry structure of brain images which is suitable for quick analysis of a large number of medical image data.This method consists of two parts: offline part and online part.In offline part,feature region and non feature region,feature region point set and non feature as well as reference point set and confusing point set are defined.Firstly,the feature region point set is obtained by our designed offline auxiliary tool.Secondly,we do filtering and fusion on the obtained set to get the optimized feature point set;in online part,we first use bi-directional matching method to get matching point set by matching to feature point set generated in offline part.And the matching point set is pruned to be optimized.Finally,the i ML of input brain CT image is fitted by the optimized matching point set.(2)We present a new medical image retrieval model based on an iterative block coding tree(BC-Tree).The corresponding methods for coarse-grained and fine-grained similarity matching are also proposed.Moreover,a multi-level index structure is designed to enhance the retrieval efficiency.Combined with the k NN classification method,using BC-Tree to optimize the classification efficiency,we get a robust classifier for large scale brain CT image.(3)We designed a multi-stage classified visualizing system based on the interaction model for symmetric two-dimensional medical image.This system visualizes the whole classifying process by a simple human-computer interaction and a display of real-time graphics,images,and animation.And during the visualizing process this system applies a simplified interacting strategy that will simplify the interactive process of feature extraction and,reduce the time of training and classification,when the amount of the processed image increases.