Diagnosis of Alzheimer's disease from MRI images of the human brain by algorithms based on machine learning

Abstract:

Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by a decline in memory and cognitive abilities. Symptoms include gradual memory loss, changes in personality, difficulties in reasoning and decision-making, and a decline in performing daily activities. As the disease advances, individuals lose the ability to manage their own lives. Brain imaging using MRI provides high-resolution images of brain structures, offering various insights into brain tissues and structures, including cerebrospinal fluid (CSF), white matter (WM), and gray matter (GM). Changes in these structures can serve as indicators of brain disorders, including Alzheimer's. In recent decades, with advancements in technology and significant progress in traditional machine learning, algorithms and models for disease detection have seen increased use. This thesis introduces several new methods for classifying Alzheimer's disease using traditional machine learning algorithms. In essence, a diagnostic tool has been developed to accurately detect Alzheimer's using brain images. Using the Kaggle database, three different methods for Alzheimer's disease classification are proposed. The first and second methods are based on dimensionality reduction. In the first method, after MRI image retrieval, texture-based features are extracted using Gabor transform algorithms at angles of 10, 15, 25, and 35 degrees. Subsequently, dimensionality reduction is performed using the Locally Sensitive Discriminant Analysis (LSDA) algorithm. In the second method, MRI images are segmented into three classes (WM, GM, CSF) using fuzzy C-means clustering (FCM) and particle swarm optimization (PSO). The subsequent steps are similar to the first method. In the third method, after MRI segmentation into WM, GM, and CSF classes, texture features are extracted using Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) algorithms. Data balancing is then achieved using the Synthetic Minority Over-sampling Technique (SMOTE) algorithm. Finally, each method utilizes different classifiers for binary and four-class classification. The classification results show that in the existing methods, the first method achieves accuracies of 92.5% and 91.19% for binary and four-class scenarios, respectively. The second method achieves accuracies of 92.18% for binary classification and 93.51% for four-class classification. The third method outperforms others with accuracies of 99.93% for binary and 99.98% for four-class classification scenarios. In conclusion, the third method demonstrates superiority in accuracy compared to other methods, providing a robust framework for Alzheimer's disease classification using traditional machine learning algorithms applied to MRI images.