Alzheimer?�?S Disease Prediction Using Deep Learning

Abstract

Alzheimer?�?s Disease (AD) represents a significant global health challenge due to its progressive nature and the absence of effective treatments. This study aims to enhance the early detection and classification of AD using deep learning techniques, specifically focusing on MRIdata. By addressing the limitations of existing diagnostic models, particularly the issues relatedto data imbalance, we developed and evaluated two model architectures: a custom Convolutional Neural Network (CNN) and a modified VGG16 model. Our methodology involved the collection and pre-processing of MRI datasets from publicly available sources, ensuring rigorous data handling to maintain quality and relevance. Data augmentation techniques were selectively applied to address class imbalance, notably using the Synthetic Minority Over-sampling Technique (SMOTE). The custom CNN architecture was designed tohandle imbalanced datasets effectively, achieving a test accuracy of 99.22%, with precision, recall, and F1 scores all at 99.22%. This model demonstrated robustness and high predictive accuracy across different stages of AD, as validated by its confusion matrix. For balanced datasets, the modified VGG16 model, pre-trained on Image Net and fine-tuned for our specific task, was utilized. This model achieved outstanding performance with an accuracy of 97.98%,precision of 98.12%, recall of 97.82%, and an AUC of 99.86%. The balanced dataset allowed the VGG16 model to leverage its depth and feature extraction capabilities fully, resulting in ahighly effective classification tool for AD stages. Comparative analysis indicated that the modified VGG16 model outperformed the custom CNN in scenarios with balanced data, underscoring the importance of dataset preparation in achieving optimal model performance. The findings from this study highlight the critical role of advanced deep learning models and meticulous data pre-processing in enhancing diagnostic accuracy for Alzheimer?�?s Disease. This research not only contributes to the field of medical imaging and diagnostics by demonstrating effective strategies for adapting pre-trained models but also sets a new benchmark for Alzheimer's disease classification using MRI images. Future work will focus on acquiring more diverse datasets, exploring additional deep learning architectures, and implementing these models in real-world clinical environments to further validate their utility and effectiveness.