Abstract
Alzheimer's, or so-called dementia, is one of the types of diseases that affects brain cells and causes memory loss, difficulty in thinking, and forgetfulness. Thus far, there is no effective treatment for AD, but treatment could be helpful in impeding the progression of the disease. Therefore, early AD diagnosis is effective in limiting the disease from progressing to advanced and dangerous stages. Physicians and radiologists face difficulties in diagnosing healthy nerve cells from soft tissue, and it requires substantial expertise and a long time to decipher the MRI images. Thus, artificial intelligence techniques can play a key role in diagnosing MRI images for early detection of AD. In this study, four proposed systems with different methodologies and materials for tracking the stages of AD development are presented. The first proposed system is to classify a data set using artificial neural networks (ANNs) and feed-forward neural networks (FFNN) based on the features extracted in a hybrid manner by using a combination of Local Binary Pattern (LBP), Discrete Wavelet Transform (DWT), and Gray Level Co-occurrence Matrix (GLCM) algorithms. The second proposed system is to classify the data set using two deep learning models - ResNet-18 and AlexNet - that are pre-trained based on deep feature map extraction. The third proposed system is to diagnose the data set using a hybrid technology between ResNet-18 and AlexNet models to extract feature maps and machine learning (SVM) to classify feature maps. The fourth proposed system diagnoses the data set using ANN and FFNN algorithms based on the hybrid features of ResNet-18 and AlexNet deep learning models and traditional algorithms (LBP, DWT, and GLCM). All the proposed techniques achieved superior results in the diagnosis of MRI images for early detection of AD. The FFNN algorithms based on the hybrid features extracted by ResNet-18 with features extracted using traditional algorithms achieved an accuracy of 99.8%, precision of 99.9%, sensitivity of 99.75%, specificity of 100%, and AUC of 99.94%.