ALZHEIMER’S DISEASE PREDICTION BY USING DEEP STACKED ENSEMBLE MODEL ENHANCED WITH SQUEEZE-AND-EXCITATION ATTENTION MECHANISM

Abstract

Context. Alzheimer’s disease (AD) is a progressive, neurological degenerative disease causing memory loss, impaired cognition, and dementia. Timely identification of AD is crucial for the provision of effective treatment and intervention. Magnetic Resonance Imaging (MRI) has also become a critical tool in understanding the structural changes in the brain that occur during Alzheimer’s development. Nonetheless, the manual processing of MRI scans is time-consuming, subjective, and susceptible to human error. As such, there is increasing demand for automated and precise diagnostic technology that can support clinicians in the earlier detection and staging of Alzheimer’s disease based on medical imaging data.
Objective. The present study focuses on developing and evaluating a deep learning-based stacked ensemble model for the classification and staging of Alzheimer’s disease brain MRI scans. The primary objective is to improve the diagnosis accuracy and reliability through a combination of the strengths of several pre-trained convolutional neural network (CNN) architectures, combined with sophisticated attention mechanisms and meta-learning techniques.
Method. The proposed approach utilizes a deep stacked ensemble learning framework composed of three well-performing CNN architectures: MobileNetV2, ResNet50, and DenseNet121. These models are pre-trained on the ImageNet dataset, benefiting from robust feature extraction capabilities. To further improve their performance, each CNN model is enhanced with a Squeeze-andExcitation (SE) attention module, which adaptively recalibrates channel-wise feature responses, emphasizing important features while suppressing irrelevant ones. The extracted high-level features from all three SE-augmented CNNs are then concatenated and fed into a meta-learner consisting of fully connected layers. This meta-classifier incorporates dropout and batch normalization techniques to prevent overfitting and improve generalization. The overall architecture is trained and validated on a dataset of brain MRI images categorized into different stages of Alzheimer’s disease, including normal control, mild cognitive impairment, and various stages of dementia.
Results. The experimental evaluation demonstrated exceptional performance, achieving an Accuracy of 99%, a Precision of 99%, a Recall of 98%, and an F1-score of 99%. These metrics indicate the model’s strong predictive capability and reliability in
distinguishing between different stages of Alzheimer’s disease.
Conclusions. The experimental outcomes highlight the effectiveness and robustness of the proposed deep stacked ensemble model in the automated diagnosis and staging of Alzheimer’s disease using MRI scans. The integration of multiple CNNs with
attention mechanisms and meta-learning significantly enhances classification performance. These findings suggest that the model can serve as a reliable decision-support system for neurologists, aiding in early diagnosis, timely intervention, and improved patient outcomes in clinical settings.