Study On The Early Diagnosis Methods For Dementias Based On Asymmetry Features Of Brain MR Images

Dementia refers to chronic acquired progressive intelligent disorder syndrome, the key of prevention and cure is early detection and intervention. Alzheimer's disease(AD) and vascular dementia(VD) are the most common types of dementia. Mild cognitive impairment(MCI) and vascular cognitive impairment(VCI), which are transitional states between normal aging to AD and VD, will convert to dementia in certain probabilities. Besides, during the pathologic process of dementia, the asymmetry of the brain will change. Therefore, to explore a universal method for early diagnosis of dementia based on brain asymmetry is of great importance.Magnetic resonance imaging(MRI), which has advantages of no invasive, high spatial resolution and radiation-free, can quantitatively reflect the brain changes in the structural and functional of different tissues and characterize metabolite concentrations, has been widely used in the early diagnosis of AD. In this paper, we extracted various asymmetric features from MR images, designed integrated model of wrapper feature learning and classification based on machine learning, and performed feature learning and classification to obtain higher and more stable accuracy for diagnosis, which provided ideas for early dementia diagnosis based on asymmetric image features. The main works of this paper are as follows:Wrapper feature selection and classification model for early diagnosis of AD and VD was researched and implemented. Through chain-like agent genetic algorithm(CAGA), global feature selection was performed, in which classification accuracy of support vector machine(SVM) was work as fitness value for CAGA, in order to select the optimal feature subset, as well as to reduce the feature dimension and improve the accuracy of classification.For AD,the classification ability of individual and small amount of asymmetry features were tested, respectively, and the distribution of excellent features was also tested. By comparing the capability of feature classification before and after feature selection, the effectiveness of the wrapper feature selection and classification model was verified. And the distinguishing areas and features for classification in the process of normal aging to the conversion of MCI to AD were analyzed.For VD, the wrapper feature weighting and classification model was also researched and implemented. Through the CAGA, global feature weighting was performed, in which classification accuracy of support vector machine(SVM) was fitness value for CAGA, in order to obtain the optimal feature weights and the accuracy of classification. The perfomances of feature selection and feature weighting were compared.This paper studied brain MR image asymmetric features for AD and VD early diagnosis based on machine learning, obtained a high accuracy of classification, and researched the role of asymmetric features, in-depth understanding of the mechanism, to provide new ideas and methods for the early diagnosis of AD and VD, layed a certain theoretical and method foundation for clinical and practical development.