مريم شريفي رناني

This thesis aims to improve early detection systems for Alzheimerʹs disease through the analysis of MRI images using deep learning models. Alzheimerʹs disease, as one of the most common neurological disorders, requires early diagnosis to prevent its rapid progression. In this study, the proposed model was designed based on the EfficientNet-B0 architecture an‎d the CBAM attention mechanism. EfficientNet-B0 was chosen due to its low number of parameters an‎d high generalization capability, while the addition of the CBAM module allowed the model to better focus on important features in MRI images. The MRI data used in this study were obtained from the reputable Kaggle platform an‎d included four distinct classes: Non-Demented, Very Mild Demented, Mild Demented, an‎d Moderate Demented. The dataset was divided into three subsets: 70% for training, 20% for testing, an‎d 10% for validation. Additionally, preprocessing techniques such as normalization an‎d image resizing were applied to prepare the data for training. The proposed model was trained with a learning rate of 1e-4 using the Adam optimizer, an‎d the Early Stopping technique was employed to prevent overfitting. The eva‎luation results demonstrated that the proposed model outperformed baseline models such as ResNet, DenseNet, an‎d VGG in metrics such as accuracy, recall, F1-Score, an‎d precision. The model achieved a test accuracy of 97.54%. Furthermore, the proposed modelʹs compact size (3.33 MB) an‎d the total number of parameters (873,356) indicated its efficiency an‎d suitability for deployment on resource-constrained systems. In conclusion, this study presents a lightweight, accurate, an‎d reliable model, making a significant contribution to the early detection of Alzheimerʹs disease using deep learning. The findings of this research can serve as a foundation for future studies in this domain.

يادگيري عميق , شبكه عصبي كانولوشن , مكانيسم توجه

Deep Learning , convolutional neural networks , attention mechanism

Sharifi

Dr. Mirzakuchaki