Prognostic Classification Study Of Alzheimer's Disease Based On Support Tensor Machine Algorithm And T1-weighted MRI

Alzheimer's Disease(AD)is a neurodegenerative and irreversible disease with occult onset,which is more common in people over the age of 65,is one of the most common type of dementia all over the world.In 2016 world AD report,the global number of people who suffered from dementia has reached 47 million,including 50%-75%for AD patients.At present,the number of people with AD in China has ranked first in the world,and China is also one of the fastest growing countries in the world.However,the diagnosis and treatment of AD has a strong contrast with the incidence rate.With the global AD and other types of dementia,only 22%of the patients received a diagnosis.In China,the proportion is lower,49%of cases were mistaken for natural aging,only about 21%of the patients received a standardized diagnosis,and only about 19.6%received medical treatment.But on the current level of medical treatment,AD is an incurable disease.Therefore,it is very important to early diagnosis and intervention treatment to AD.Mild Cognitive Impairment(MCI)is a preceding state that previously proposed for Alzheimer's disease,and it's an intermediate state between normal aging and dementia.MCI can be used as "predictor" of AD,and it can be completely cured.So if we can find this state as early as possible and to give appropriate intervention treatment,it can delay the progress of AD.Thus,correct diagnosis of AD,especially at the early stages of the diagnosis and treatment of mild cognitive impairment is very important for the prevention,early detection and treatment intervention to AD.In order to identify Alzheimer's disease and mild cognitive impairment,this paper presents a diagnosis method based on support tensor machine(STM)classifier with the gray level features of T1 weighted MRI brain gray image.In this classifier,three dimensional(3D)gray matter image is used as the model input,and the weight vector of each mode is trained by STM iterative algorithm for classification.T1 weighted MRI three-dimensional(3D)brain images from 70 AD patients,112 MCI patients(included patients were converted to AD during follow-up,MCI-C:MCI Converters and patients were not converted to AD during follow-up,MCI-NC:MCI Non-converters)and 70 NCs(normal controls)are collected.The third-order tensors are obtained by extracting image intensity of each voxel of gray matter.The size of third-order tensors is 95×119×102.Tensor principal component analysis(TPCA)is used to obtain the low dimensional principal component tensor of the three order grey matter tensor,and the principal component tensor is used as the input of the classifier based on STM(STM-TPCA).Tensor independent component analysis(TICA)is also used to obtain the independent component tensor of the three order grey matter tensor,and the independent component tensor is used as the input of the classifier based on STM(STM-TPCA).Considering the feature information of three order tensor is redundant,therefore,we adopt recursive feature elimination method(Recursive feature elimination,RFE)combined with support tensor machine(Support Tensor Machine,STM)to vector the three order tensor data and select useful features for each mode of the three order tensor to get the best subset of features as the input of the classifier(STM-RFE).Finally,the classification of four groups,such as AD&NC,MCI&NC,AD&MCI,MCI-C&MCI-NC,is implemented by using 10-fold cross-validation method.This classification model is trained by 7-fold cross-validation method.For the classification of AD and NC,the accuracy rate reaches 91.19%(sensitivity 92.86%,specificity 89.52%);for MCI and NC,the accuracy is up to 83.15%(sensitivity 91.67%,specificity 69.52%);for AD and MCI,the accuracy is 82.23%(sensitivity 65.71%,specificity 92.56%);for MCI-C and MCI-NC,the accuracy is 77.08%(sensitivity 77.38%,specificity 76.79%).In addition,basic information(age,gender and education level)and cognitive scores((Mini-Mental State Exam,MMSE score;Alzheimer's Disease Assessment Scale-cognitive subscale,ADAS-cog score)are combined with the third-order tensor for classification.The results show that the performance of the classification model can be further improved by combining the basic information and cognitive scores,and the classification results are better than Shen's studies and Willette's studies,which are also combining multi-modal data.The experimental results show that the diagnosis method based on support tensor machine(STM)classifier with the gray level features of T1 weighted MRI brain gray image is an effective method for diagnosis of AD;and basic information,cognitive scores and MRI gray matter image are compatible,which has a very good complementary effect.In the process of the experiment,we found that TPCA and RFE methods is running very slow,this is because the number of each mode of feature tensor are relatively high(95×119×102),the high dimension features greatly increase the time of feature extraction and feature selection.Therefore,we consider the texture features tensor(12×13×4)as the input of the STM(STM-Texture)based on the gray level co-occurrence matrix(GLCM)of 13 directions and 4 kinds of distance.This method reduces the number of features of each mode,thus improving the speed of the algorithm.The results show that the tensor method combined with texture features tensor(12×13×4)based on the gray level co-occurrence matrix(GLCM)can maintain the superiority of the original classification effect,and reduce the running time of the model,and improve the robustness of the model.