Enhanced Alzheimer's Disease Classification: A Stacked Model Fusion with Brain MRI Imaging

Alzheimer's disease (AD) is still a major global health issue that requires sophisticated staging and classification techniques. This paper introduces a novel approach to AD categorization, employing a stacked network architecture and feature extractions from InceptionV3, EfficientNetB0, and MobileNet models. The primary objective is to address the challenge of imbalanced datasets while minimizing the risk of overfitting. The proposed method utilizes brain MRI scans, extracting features and incorporating a Global Average Pooling (GAP) layer to improve the identification of critical features that correspond to AD phases. Combining the retrieved features results in a single set that helps with stage discrimination, followed by finishing via a SoftMax classifier. Model evaluation and validation are essential steps in determining the ideal model to minimize the risk of overfitting. A comprehensive evaluation is ensured by cross-validation (CV) using a 3-fold technique, which shows that the suggested Stacked Model has a better accuracy of 94.58%, closely followed by EfficientNetB0 at 90.70%. Both InceptionV3 and MobileNet exhibit competitive results, with accuracy rates of 89.46% and 87.60%, respectively. By presenting a strong approach to AD classification, this study advances the field of early detection and staging of AD by demonstrating the effectiveness of the Stacked network in identifying complex features that are indicative of different stages of AD progression within brain MRI datasets.