Comparison Between U-Net and Googlenet for Alzheimer's Disease Detection and Analysis Using MRI Images

Abstract

Alzheimer’s disease (AD) is a brain disorder that causes gradual memory loss and cognitive decline. Early detection is important because it helps doctors manage the disease better. MRI scans are often used to find early signs of Alzheimer’s, and deep learning models, like Convolutional Neural Networks (CNNs), have become useful tools for analyzing these images. This paper focuses on a comparative analysis of two prominent deep learning architectures, U-Net and GoogLeNet, in the context of Alzheimer’s disease research using MRI images. U-Net, known for its encoder-decoder architecture with skip connections, is widely used for biomedical image segmentation, such as hippocampal segmentation in Alzheimer's disease. It excels in tasks requiring precise localization of brain structures. Conversely, GoogLeNet, designed for image classification, employs Inception modules for multi-scale feature extraction, making it suitable for distinguishing between Alzheimer’s and healthy brain images. The authors developed a Convolutional Neural Network (CNN) to classify MRI images into four categories: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented. The model was trained on a dataset comprising 6,400 images gotten from Kaggle, and was split into 70% for training, 24% for validation and 6% for testing. The training process involved several data preprocessing techniques such as, normalization, and data splitting. U-Net and GoogLeNet, a pre-trained model, served as the foundation for the network, which was fine-tuned for the task of Alzheimer's stage classification. The model was optimized using the Adam optimizer and trained for 20 epochs with categorical crossentropy as the loss function. The results indicated a training accuracy of 85% for U-Net while GoogLeNet indicated a training accuracy of 93.50%. Despite moderate success, the gap between training and testing performance suggests room for improvement, particularly in addressing class imbalance and fine-tuning the model. The study concludes that while the deep learning approach shows promise in aiding Alzheimer's diagnosis, further enhancements such as advanced data augmentation, model refinement, and expanded clinical application are needed to fully realize it’s potential.