![]() | Only 14 pages are availabe for public view |
Abstract The theoretical and practical advances on Artificial Intelligence in last 20 years have permeated and benefited a spread of fields, like medicine, tourism, education, entertainment, among others. In medical field, AI has proven the potential to aid in the detection and diagnosis in Alzheimer’s Disease (AD). And this is done through lesion detection, cells and organs segmentation, automatic conversation, among others; the supervision of patients remotely, the discovery of medications, robot-assisted surgeries, among others. With the development of medical imaging techniques, neuroimaging plays a major role in the diagnosis of Alzheimer’s Disease encompasses Computer Tomography (CT), Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), Functional Magnetic Resonance Imaging (FMRI), and Single-Photon Emission CT. MRI is used to analyze structural changes caused by AD because of its ease of accessibility. This thesis is concerned with applying Artificial Intelligence techniques in AD diagnosis. The main objective is to introduce various intelligent techniques for diagnosing AD. To achieve this objective, the thesis first provides a survey on the most popular artificial intelligence algorithms used in detecting and classification AD highlighting their strengths and weakness. Then, the thesis presents two methodologies based on utilizing Machine Learning algorithms including Deep Learning, Convolutional Neural Networks and Sparse Autoencoder (SAE). Initially, the thesis proposed intelligent model to predict AD with a deep 3D Convolutional Neural Network (3D CNN), which can categorize diseased brain from the healthy brain based on MRI scans. This model achieved 96.5 for training data and for test data reached 80.6%. According to this investigation, the shift and scale invariant features retrieved by 3D-CNN followed by deep learning classification are the most powerful approach to identify clinical data from healthy data in MRI. Furthermore, the second model decomposes into two steps: First, in order to apply unsupervised feature leaning, training sparse autoencoder (SAE) model was achieved. In this model, the thesis spot the light on evolving a SAE model to discover the most operative features from the AD dataset. Second, developing a 3D-Convolutional Neural Network (3D-CNN) to differentiate between the health status and diseased status MRI scan of the brain. This model reached to 93% accuracy for the training data, 94% accuracy for validation data and 87, 87% accuracy for testing data. |