Abstract

The present research explores the early identification of Alzheimer's disease (AD) phases, encompassing Mild Cognitive Impairment (MCI), a transitional stage potentially facilitating disease prevention efforts. This investigation explores the diagnosis of Alzheimer's disease (AD) stages through the application of a multiclassification-based deep learning approach, in contrast to the predominant focus on binary classification methods for AD identification in existing research. The study utilizes the Alzheimer's Disease Neuroimaging Initiative (ADNI) clinical dataset, which encompasses over 2000 samples and exhibits an imbalanced distribution, with AD or Dementia representing the minority class. Deep Ensembled Learning is applied to the dataset with seven selected biomarkers to diagnose the disease stage through multiclassification. The ensemble approach is effective in enhancing reliability and demonstrates improved diagnostic performance for the AD stage, which belongs to the minority category. While the majority of prevalent studies utilize the AUC score to measure the performance of AD diagnosis using binary classification, this study employs both the F1 score and AUC score in the multiclassification of AD stages. The multiclassification for diagnosis yielded F1 scores of 88% for Cognitive Normal (CN), 86% for Mild Cognitive Impairment (MCI), and 86% for Alzheimer's Disease (AD) stage detection. The overall accuracy obtained is 87%, while the Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) is 91% for CN, 87% for MCI, and 91% for AD. Performance of diagnosis of AD / dementia stage that belongs to minority samples demonstrate an increase by 6% when compared with prior research [1]. The utilization of the established ADNI dataset and an ensemble approach has enhanced the reliability of the results. F1 score and AUC are effective measures, as the dataset is imbalanced. The utilization of the essential clinical biomarkers for accurate AD stage diagnosis is both efficient and effective.

Statistics