Research On Feature Selection Method Of MRI Structure And The Application Of AD Early Diagnosis

In recent years, the aging problem in China is becoming more and more serious, patients with Alzheimer’s disease(AD) and mild cognitive disorder(MCI) is also increasing, MCI is the early process of Alzheimer’s disease, the disease is a neurodegenerative disease, the current medical treatment has yet to fully cure the disease, but is if the early diagnosis and drug intervention can alleviate the condition of development. Current methods of diagnosis mainly rely on experience of reading the doctor through the analysis of the patient’s brain structural magnetic resonance imaging(MRI) image to determine the patient’s condition, such a method is not only time consuming power consumption, and there is a very strong subjectivity, may lead to misdiagnosis. Computer aided diagnosis will help to improve the diagnostic accuracy and reduce the workload.With the development of machine learning techniques, feature selection method has been applied in various fields of study, feature selection method has achieved certain effects in the field of medical auxiliary diagnosis, but due to the large number of medical imaging features, existing methods cannot be applied directly to medical diagnosis.To this end, the paper proposed heuristic experience based search(HS-EJ) feature selection model, to a large amount of information in the MRI data were selected to extract useful features for aided diagnosis of MCI and AD. The main research work is as follows:(1) Collected MCI, AD patients and healthy people(NC) of the brain MRI image structure, and carried on the pretreatment, extraction of three groups of subjects in gray and white matter and cerebrospinal fluid volume characteristics.(2) Created the model of HS-EJ feature selection: first, the original volume data were significant analysis and logistic regression analysis, on the basis of the significant level and the regression coefficient were sorted, feature elimination poor classification effect; then according to the sequential forward search strategy, followed by the classification features into SVM classifier, the classification accuracy of the selected feature set as the highest optimal feature subset; by contrast, from two groups of optimal feature selection feature better set.(3) Compared with the HS-EJ feature selection model and significance analysis of filtering model and logistic regression analysis model, SVM filter classifier packaging model, principal component analysis model of feature selection efficiency.(4) By using multiple linear regression to get the best feature subset for further reduction, will merge into a multi feature feature classification results using ROC curve showed Compared combination features with other single feature.The number of features and classification of the classification and comparison of HS-EJ feature selection model and other feature selection models are accurate, concluded: HS-EJ feature selection significantly improves the classification accuracy rate of two two NC-MCI-AD, has a certain clinical significance for early diagnosis of MCI and AD.