Gaussian Prediction Models Of Early AD Based On Contourlet Texture Features

Background: Background dementia is one of the common diseases in the elderly. Globally, 1 new cases of dementia are added every 7 seconds in the world, by 2050. Alzheimer’s disease(AD) is a common dementia in a, in China over the age of 65 Elderly AD prevalence rate of 4.8%(currently in China about 600 million AD patients), from the onset of clinical symptoms to diagnosis of first time an average of more than a year, so at the time of diagnosis mostly have missed the best treatment phase. So it is one of the hot and difficult problems at home and abroad to study the early diagnosis of AD.For the early diagnosis of AD, it is usually used to find the suitable biomarkers as the basis for the diagnosis. At present, there are 5 kinds of biological markers, including brain atrophy, which is presented by magnetic resonance imaging(MRI). At home and abroad, the researchers have used brain atrophy to predict the early stage of AD. The texture and model are reported, but the texture and model are relatively backward. Early in the project group were applied to the first and the second generation wavelet transform and gray symbiotic matrix method to establish lung cancer prediction model(including multi level model, support vector machine, Lasso regression, decision tree, random forest, artificial neural network, gradient boosting and recently neighbor classification), which supports vector machine can obtain better prediction results. At present, the research of the early AD prediction model for the extraction of texture at home and abroad has the following limitations: the selected texture parameters are less, and the general combination of brain morphological parameters. Forecasting models are artificial neural networks and so on, and the sample size is small.Objective: An early prediction model was established based on the texture value parameters and morphological parameters of brain MRI images, which was found to increase the recognition rate of early AD and the transformation of mild cognitive impairment(MCI).Methods: The study collected 299 cases of brain MRI images, including 58 cases of AD patients, 147 cases of MCI patients, 94 cases of normal group, and each case has more than two years of follow-up.In order to compare support vector machine, Gaussian process regression model and partial least-squares regression model of three methods for early AD prediction results of the pros and cons of using bootstrap simulation data, model predictions.Using the region growing method to segment the hippocampal region from the coronal MRI images, the segmentation of the hippocampal area was processed by Contourlet transform and the 14 texture parameters were calculated by the gray level co-occurrence matrix.147 cases of MCI were used as validation data sets, respectively. The prediction models of AD and MCI were established respectively, and the corresponding evaluation indexes were calculated. Comparison of several models for predicting the effect of MRI data.Through consulting literature websites and advisory imaging specialists in AD group, MCI group and the normal group of demographic data, the living environment factors and neural weight scale, based on the texture values and demographic characteristics, history of illness, brain morphological characteristics of multi dimension data set, the prediction model was established.By surveys, we got demographic features, environmental features, genetic features and morphological features, and combined textural features and other features to establish prediction models as an ancillary diagnostic tool.Results:1 based on the univariate analysis showed that the age of patients in the AD group, MCI normal group in 3 cases. There was no statistically significant difference(Z=3.45, p=0.64), were years of education in the difference between the 3 groups was not statistically significant(Z=6.45, p=0.85). Scale, MMSE, CDR-SB, the difference was statistically significant in three groups in ADAS. 8 cases of disease, history of illness between the 3 groups was statistically significant; brain volume differences morphological parameters, 184 morphological volume parameters were statistically significant. There are 379 aspects of texture, texture parameters the difference was statistically significant in three groups.2 the establishment of support vector machine based on Contourlet transform, Gauss, partial least squares regression model. Through the evaluation index, the prediction model of partial least squares regression model of the best prediction results(sensitivity 0.90, specificity of 0.78, area under the ROC curve 0.90), the process of Gauss(sensitivity 0.84, specificity of 0.75, area under the ROC curve0.84).3 through the factor analysis, show several texture value parameters have obvious contribution to the prediction of the model. For Standard Deviation, Homogeneity, Energy, Inertia, Inverse, Difference, Moment, Correlation, Difference Mean and Sum Entropy.The prediction accuracy of 4 partial least square model of MCI reached 85.5%, Gauss 82.2%, Support vector machine is 80.3%.Conclusions1 the prediction model can accurately predict the early AD and MCI conversion, and three models, the best effect of partial least squares. 2 of the value added texture has a positive effect on early prediction of AD.