Deep Learning Approaches with Explainable AI for Differentiating Alzheimer Disease and Mild Cognitive Impairment

TL;DR

This work tackles early and accurate differentiation of Alzheimer’s disease (AD) and Mild Cognitive Impairment (MCI) using structure MRI. It proposes a hybrid ensemble framework that fuses three pretrained CNNs (ResNet50, NASNet, MobileNet) through a two-stage transfer learning process (feature freezing followed by fine-tuning) and combines their outputs via both weighted averaging and a stacked meta-learner, enhanced by Gradient-weighted Class Activation Mapping (Grad-CAM) for interpretability. On the ADNI dataset, the approach achieves state-of-the-art performance, exemplified by AD vs. MCI accuracy of $99.21\%$ and MCI vs. NC accuracy of $91.02\%$, with ROC-AUC in the high 0.98–1.00 range, while providing heatmaps that localize discriminative neuroanatomical regions such as the hippocampus. This methodology offers a scalable, robust, and interpretable framework for clinical decision support in neurodegenerative diagnostics, and it points to future extensions using domain adaptation and multi-modal data to further enhance generalization and utility in diverse clinical settings. Specifically, the ensemble probability can be expressed as a convex combination $\hat{p} = \boldsymbol{\alpha}^\top \mathbf{p}$ and refined through a meta-learner $g(\cdot)$ trained on out-of-fold predictions, enabling non-linear fusion that improves discrimination across AD, MCI, and NC groups.

Abstract

Early and accurate diagnosis of Alzheimer Disease is critical for effective clinical intervention, particularly in distinguishing it from Mild Cognitive Impairment, a prodromal stage marked by subtle structural changes. In this study, we propose a hybrid deep learning ensemble framework for Alzheimer Disease classification using structural magnetic resonance imaging. Gray and white matter slices are used as inputs to three pretrained convolutional neural networks such as ResNet50, NASNet, and MobileNet, each fine tuned through an end to end process. To further enhance performance, we incorporate a stacked ensemble learning strategy with a meta learner and weighted averaging to optimally combine the base models. Evaluated on the Alzheimer Disease Neuroimaging Initiative dataset, the proposed method achieves state of the art accuracy of 99.21% for Alzheimer Disease vs. Mild Cognitive Impairment and 91.0% for Mild Cognitive Impairment vs. Normal Controls, outperforming conventional transfer learning and baseline ensemble methods. To improve interpretability in image based diagnostics, we integrate Explainable AI techniques by Gradient weighted Class Activation, which generates heatmaps and attribution maps that highlight critical regions in gray and white matter slices, revealing structural biomarkers that influence model decisions. These results highlight the frameworks potential for robust and scalable clinical decision support in neurodegenerative disease diagnostics.

Figures (4)

• Figure 1: Hybrid deep ensemble learning framework with XAI using ResNet-50 and a dense meta-learner for Alzheimer’s Disease classification. Grad-CAM highlights key brain regions in both coronal and horizontal MRI sections distinguishing healthy control from Alzheimer’s patients.
• Figure 2: Workflow of the proposed Hybrid Deep Ensemble model for early diagnosis of Alzheimer’s disease. The pipeline begins with the preprocessing and slicing of MRI images (gray matter and white matter)
• Figure 3: ROC curve comparison for eResNet50, eNasNet, eMobileNet, eEnsemble, and the proposed Hybrid Ensemble with 10 fold CV.
• Figure 4: Grad-CAM heatmaps for three representative brain MRI slices. Top: original T1-weighted MRI images for No AD, MCI, and AD. Bottom: corresponding Grad-CAM overlays from the hybrid ensemble network highlighting class-discriminative regions.