Study On The Diagnosis Of Alzheimer S Disease Based On Radiomics On Tensor Space

Alzheimer’s disease(AD),a progressive and neurodegenerative disease,is the major common cause of dementia associated with aging.Early symptoms include memory disorders,cognition dysfunction and language barrier and one of the main pathologic features of AD is neuronal loss with consequent brain atrophy.AD affects the normal life of patients and brings heavy burden of mental and economic to families and society.Mild Cognitive Impairment(MCI),a widely used term for cognitive problems that do not fulfill the criteria for dementia,refers to a transitional phase between normal cognitive function and clinically probable AD.An in-deep study of MCI can screen out the high-risk groups of AD.Therefore,early diagnosis of potential AD is crucial for the adoption of therapeutic strategies and able to slow the progression of the disease.The emergence of magnetic resonance imaging(MRI)provides a useful method for the diagnosis of brain disease and the application in Alzheimer’s disease has made great progress.Magnetic resonance imaging has an intrinsic third-order tensor structure.Traditional vector-based machine learning methods unfold the 3D images as vectors to carry on the modeling,which break the natural 3D structure of data so that some useful information underlying the neuroimaging data is missing.In other case,the dimensionality of the vectorization data is usually high.High space dimensionality may induce small sample size problems and the curse of dimensionality for medical data with small training sample size,leading to overfitting and decreasing classification accuracy.Tensor space model was proposed to solve the problem above.The tensor space model uses high order imaging with an intrinsic high order tensor structure as input for modeling,eliminating the need for handcrafted imaging features,feature learning,or image vectorization.In recent years,great progress has been made in the research of AD prediction based on brain gray matter images.However,the study on white matter images is almost blank.White matter abnormalities indicate the early neuropathological events of AD and play an important role in the diagnosis of AD.Therefore,a novel classification method based on radiomics on tensor space is proposed in this paper.The T1 weighted MR brain white images were used as input on tensor space and then recursive feature elimination(RFE)method coupled with Support Tensor Machine(STM)was used to select the optimal features subset for classification using the STM-based classifier.The proposed algorithm performed the classification on four cases including the patients of Alzheimer’s disease and Mild Cognitive Impairment(including patients were converted to AD,MCI-C;and patients were not converted to AD,MCI-NC)and normal controls(NC),10-fold cross validation was employed to assess the classification performance then.In terms of AUC(Area Under receiver operating characteristic Curve),classification accuracy,sensitivity and specificity,the case AD vs NC archived 91.43%、89.35%、87.14%、95.71%respectively;the case AD vs MCI archived 76.37%、74.73%、71.85%、75.27%respectively;the case of MCI vs NC archived 78.57%、76.37%、78.02%、74.22%respectively;and the case MCI-C vs MCI-NC archived 81.25%、78.57%、77.36%、80.24%respectively.We combined the basic information(age,gender,education)and cognition scores(MMSE and ADAS-Cog scores)with third-order gray scale tensor and got a better result.The experimental results indicated that the proposed algorithm is effective for the diagnosis of Alzheimer’s disease.In addition,we conducted an exploratory study on multi-feature tensors for the diagnosis of Alzheimer’s disease.Firstly,a Gray-level co-occurrence matrix(GLCM)is constructed for each combination of 13 directions and 4 distances based on the gray scale tensor.And twelve texture features were extracted from each GLCM then.We combined the gray scale tensor and texture tensor and the classification performance improved a lot comparing to the former with the single feature to some extent.Secondly,we combined with the gray scale tensor of white matter and gray-matter images on the basis of existing work and we got a good classification result.