الفهرس 
ch1Only 14 pages are availabe for public view
 |  | from | 149 |  |  |
| | | from | 149 | | |
Abstract
Alzheimer’s disease(AD)is considered a common form of dementia. Morethan 55millionsADcasesweredeclaredin2021.Asthediseasedevelops,damaged or destroyed neurons are found in brain parts. Eventually, in partsof the brain, neurons that help a person to perform essential functions ofthe body are affected, such as swallowing and walking. Activities that aperson used to perform are no longer possible for the patient with AD, such as planning events for family or participating in sports. In the final stagesof AD, the patients are bed-bound and need care. Current research cares about identification of stages of AD. AD is often divided into four generalstages : normal control(NC),early mild cognitive impairment(EMCI),lateMCI (LMCI),and AD. Symptoms are present in the last two stages wit varying degrees. The pathogenesis of AD remains not fully elucidated, and no available therapy can cure AD or completely stop disease progression. MCI is a transitional stage between cognitively normal aging and AD, and patients with MCI are more likely to develop AD than age-matched NC. Early detection of AD by screening MCI is crucial both for effective management and care strategies and for developing new drugs and measures to prevent further deterioration of the disease. So, we work in two directions as the following : genome sequence analysis, and medical images analysis. In the first direction, the researchers in genome sequence analysis face some problems and challenges. Genotyping millions of SNPs is costly. An appropriate subset of SNPs is required to represent the rest of the SNPs accurately. Identifying new genes linked to AD is essential to understand better the pathological mechanisms that lead to neurodegeneration.Geneticdiagnosticproceduresarecrucial,especiallyat early stage of AD because it can be challenging to differentiate between different neurodegenerative diseases. Also monitoring epigenetic changes and transcriptional analysis. These studies may be a favorable approach to various genetic disorders that help to understand the entire genetic mechanisms of diseases, such as AD. In the second direction, medical images are considered one of the most popular practices recently ,as their presence in digital form for building computer aided diagnosis(CAD)applications for image processing/analysis techniques. Linking different medical images to take the advantage foreachmodalityisconsideredasachallengeforaccurateearlydetectionofADand differentiate between different stages of AD. In this thesis, we propose two comprehensive framework for early AD detection and differentiate between different stages of AD which is based on integration between different modalities of data by using data fusion between SNP data and medical image. In the first framework, we propose a comprehensive framework based on an ensemble gradient boosting tree(GBT). Two featurese lection techniques are separately checked in the system :the information gain(IG)and Boruta. The two feature selection technique swere used to select the most significant AD-related genes. Filter methods measure the relevance of features by their correlation with dependent variables, while wrapper methods measure the usefulness of a subset of features by training a model on it. In this system, a GBT is used on all genetic data for ADfromneuroimaginginitiativephase1(ADNI-1)and whole-genome sequencing (WGS)datasets by using two feature selection techniques. In the ADNI 1,resultsachievedthattheGBThasanaccuracy,specicity, and sensitivity99.06%,99.00%,98.45%inthecaseofusingBorutaFS.In the case of using IG, the proposed system achieved an accuracy, specificity,and sensitivity94.87%,96.90%,98.00%.Theresultsshowthattheproposed system is preferable for early AD detection. Also, the results revealed that the Boruta feature selection is superior to information gain filter techniques. In the second framework, we propose a comprehensive framework based on a convolutional neural network(CNN)and ensemble extreme gradient boosting tree(XGBT).First, we propose an early detection of AD by using different modalities of AD brain images rather than a single image. Second, image fusion is performed using Laplacian Re-decomposition, and canonical correlation analysis(CCA)between magnetic resonance image(MRI) and Positron emission tomography(PET).Third, the feature extraction of the fusedimageisperformedbyCNNwithmodelVGG16.Finally,weuse XGB for the classification of AD. To achieve better performance and precise diagnosis results, Laplacian Re-decomposition(LRD)has been applied with XGBbasedonvisualgeometrygroup16(VGG16)as a feature extractor. Evaluating the system’s performance shows an accuracy, specificity, and sensitivity of 98.06%,94.32%, and 97.02%,respectively. Also, incase latefusion(CCA)thesystemachievedACC99.22%,specicity96.54%,andsensitivity99.54%.Thetwosystems’experimentsshowencouragingresultscompared with other systems. Also, integration between different data using decision-level fusion. we propose a decision-level fusion method thatcombines three well-known classifiers. It is used to predict the AD patienthealth for early monitoring and efficient treatment. A soft voting techniqueis used to generate the final decision result from the predictions of thesecalibrated classifiers withaccuracy99.23%.