Abstract:
"Alzheimer’s Disease (AD) is a major challenge in the medical field as the diagnosis process is time-consuming, labour-intensive, and prone to human error. The traditional diagnostic procedures heavily rely on manual analysis of MRI, which requires a high level of expertise and is vulnerable to errors leading to misdiagnoses. This can have severe consequences on patients and their care. This research addresses the critical need for an automated, accurate, and interpretable system to detect AD. Through this research, the author aims to reduce the burdens of the current diagnostic process and minimise the risks associated with diagnostic inaccuracies.
To mitigate these limitations, this research introduced a novel ensemble approach that combines the strengths of a modified AlexNet and EfficientNet B0 Convolutional Neural Networks (CNNs), improved with eXplainable Artificial Intelligence (XAI) techniques. Through this novel approach, the author aims to leverage the strengths of these Deep Learning (DL) models to improve the accuracy and overall performance of AD detection using Magnetic Resonance (MRI) Images. The adopted methodology for this research involved a blend of technical improvements for performance optimisation and a focus on interpretability. This ensures that the system is capable of providing insights into the prediction results. The utilization of XAI is expected to bridge the gap between Artificial Intelligence (AI) models and their clinical applicability.
The proposed ensemble approach with the combination of models mentioned above has not been attempted previously for AD detection and achieved a classification accuracy of 99.03% during the testing phase. This outperforms the two models that were combined to create the ensemble model, modified AlexNet and EfficientNet B0 achieved accuracies of 99% and 87% respectively. The selected XAI techniques for visual interpretations of the classification results also provided valuable insights, making the proposed approach a great step ahead in the field of AD research."
