Deep ensemble architecture with improved segmentation model for Alzheimer's disease detection.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) is the leading cause of dementia, characterized by cognitive decline impacting daily life.
• Current deep learning methods for AD detection exhibit limitations, leading to suboptimal accuracy.
• Effective early detection of AD is crucial for timely intervention and patient management.

Purpose of the Study:

• To propose a novel deep ensemble architecture for enhanced Alzheimer's disease classification.
• To address the ineffectiveness and inaccuracies of existing AD detection techniques.
• To develop a robust model for improved diagnostic performance in Alzheimer's disease.

Main Methods:

• A multi-stage approach involving preprocessing (Median filtering), segmentation (improved U-Net), and feature extraction (ISIH, MBP, Multi Texton).
• Development of a deep ensemble model, En-LeCILSTM, integrating LeNet, CNN, and improved LSTM architectures.
• Ensemble strategy combining intermediate outputs from individual models to enhance classification accuracy.

Main Results:

• The proposed En-LeCILSTM model achieved a high accuracy of 0.963 and an F-measure of 0.908.
• Performance evaluation demonstrated superior results compared to traditional AD detection methods.
• The model effectively improved the accuracy of Alzheimer's disease detection.

Conclusions:

• The novel deep ensemble architecture, En-LeCILSTM, shows significant promise for accurate Alzheimer's disease detection.
• The proposed method offers a more effective solution for the challenges in current AD diagnostic tools.
• This research contributes to advancing AI-driven diagnostic capabilities in neurodegenerative disease research.