Ensemble Convolutional Neural Networks for Alzheimer’s Disease Detection: A Systematic Review

Abstract

Alzheimer’s disease (AD) is the most prevalent form of dementia worldwide, affecting an estimated 55 million people globally and projected to triple by 2050. Timely and accurate diagnosis is critical; however, conventional clinical methods relying on neuropsychological assessments and invasive cerebrospinal fluid (CSF) biomarkers detect pathology only after significant neurodegeneration has already occurred. Magnetic Resonance Imaging (MRI) offers a non-invasive window into structural brain changes, and deep learning, particularly Convolutional Neural Networks (CNNs), has demonstrated remarkable capability in extracting discriminative spatial features from MRI volumes for automated AD classification. This systematic review investigates ensemble CNN frameworks that synergistically combine ResNet, InceptionV3, and EfficientNet architectures to overcome the inherent limitations of individual models, including sensitivity to dataset variability, architectural bias, and overfitting. Two principal ensemble strategies are critically analyzed: (1) prediction level ensembling via stacking and boosting, achieving classification accuracy up to 95%; and (2) feature-level fusion through intermediate representation concatenation, achieving accuracy as high as 99.13%. The review also examines transfer learning strategies, domain adaptation techniques, performance metrics, and the translational challenges of deploying AI based diagnostic tools in clinical settings. Findings confirm that multi-scale feature integration within ensemble CNN architectures significantly improves the robustness, generalizability, and diagnostic accuracy of non-invasive automated AD detection systems.