Research On The Aided Diagnosis Method Of Alzheimer’s Disease Based On Diffusion Tensor Imaging

Alzheimer’s Disease(AD)is the leading cause of cognitive impairment or dementia in people over the age of 65.With the increase of life expectancy,mild cognitive impairment(MCI),AD,AD-related dementia(ADRD)will be prevalent worldwide.If the symptoms of AD patients are not diagnosed in time due to being misdiagnosed by doctors or ignored by the patients themselves,treatment will be delayed and the condition will worsen.This will not only double the cost of subsequent treatment,but also cause great pain to the patient.Therefore,timely detection,accurate diagnosis and appropriate intervention of AD are very important.The genetics and etiology of AD are complex.The etiology of AD is still controversial,and AD cannot be diagnosed clinically by a single pathological feature.Therefore,doctors need to integrate AD pathological biomarkers,neuropsychological examinations and Comprehensive evaluation of imaging studies to accurately diagnose MCI and dementia due to AD.But these diagnostic procedures are not only time-consuming,but also require extensive clinical experience from neurologists.Therefore,at this stage,an efficient and automatic diagnosis method and effective image features are required to play an auxiliary role in the diagnosis of diseases by doctors.Computer Aided Diagnosis(CAD)systems refer to systems that provide information about disease assessment by interpreting medical images.Artificial intelligence can improve the performance of CAD systems by learning and analyzing the deep features of images,enabling faster and more accurate diagnosis.So far,there are many stable and high-accuracy AD diagnostic methods,but none of them can be directly used in clinic.Therefore,we consider providing interpretable imaging features while accurately diagnosing AD to assist doctors in clinical diagnosis.Diffusion Tensor Imaging(DTI),a non-invasive magnetic resonance imaging method,provides information on the directional structures of nerve bundles found in white matter and cortex.It is this link between the orientation dependence of the DTI signal and the putative underlying brain fiber orientation that provides the unique insights of diffused tract imaging.Our research work focuses on the use of machine learning and deep learning models to analyze DTI for Alzheimer’s disease diagnosis,and extract risk brain regions and risk fiber tracts.The research is mainly carried out from the following aspects:(1)Aiming at the direction error caused by the direct linear averaging of tensor channels in the construction of standard brain templates,and the linear averaging will make the junction of gray matter and white matter too smooth and reduce the resolution,this paper proposes a method that introduces four A method for constructing Gaussian templates with arity and Gaussian weighted average.Experiments show that the Gaussian template construction method proposed in this paper optimizes the linear average in the direction of the tensor and improves the resolution of the gray-white matter interface,which provides help for the subsequent construction of standard DTI brain templates.(2)In order to improve the accuracy of AD diagnosis,this paper proposes a diagnostic model framework that can simultaneously obtain the fiber structure information and anatomical structure information in the brain.The framework extracts features from brain network and combines Support Vector Machine(SVM),Random Forest(RF)and Convolutional Neural Networks(CNN)to diagnose AD to assist medical work accurate diagnosis of Alzheimer’s disease.The experimental results show that the diagnostic framework has improved in terms of accuracy,sensitivity and specificity,providing a reference for the auxiliary diagnosis of AD.(3)Aiming at the "black box" problem in the network framework,this paper proposes an FMCNN model that extracts features from fiber bundles to diagnose AD and obtains the risk probability map of fiber bundles.First,fiber tracking is performed on the DTI,and fiber bundles are tracked in different regions of interest by setting seed points.We selected fiber tracts in four regions of interest,cingulate tract,corpus callosum,uncinate tract,and white matter,as input data into three diagnostic models,CNN,MCNN,and FMCNN,respectively,and evaluated parameters through model performance.Evaluate it.Finally,the fiber probability map is output through FMCNN.Experiments show that the FMCNN proposed in this paper has improved performance compared with the original network framework,and the generated fiber high-risk probability map provides a new vision for the clinical search for the pathological features of AD,and lays a theoretical foundation for the promotion of related algorithms in clinical applications.