Background/Objectives: Alzheimer’s disease (AD), a progressive neurodegenerative disorder, demands precise early diagnosis to enable timely interventions. Traditional convolutional neural networks (CNNs) and deep learning models often fail to effectively integrate localized brain changes with global connectivity patterns, limiting their efficacy in Alzheimer’s disease (AD) classification. Methods: This research proposes a novel deep learning framework for multi-stage Alzheimer’s disease (AD) classification using T1-weighted MRI scans. The adaptive feature fusion layer, a pivotal advancement, facilitates the dynamic integration of features extracted from a ResNet50-based CNN and a vision transformer (ViT). Unlike static fusion methods, our adaptive feature fusion layer employs an attention mechanism to dynamically integrate ResNet50’s localized structural features and vision transformer (ViT) global connectivity patterns, significantly enhancing stage-specific Alzheimer’s disease classification accuracy. Results: Evaluated on the Alzheimer’s 5-Class (AD5C) dataset comprising 2380 MRI scans, the framework achieves an accuracy of 99.42% (precision: 99.55%; recall: 99.46%; F1-score: 99.50%), surpassing the prior benchmark of 98.24% by 1.18%. Ablation studies underscore the essential role of adaptive feature fusion in minimizing misclassifications, while external validation on a four-class dataset confirms robust generalizability. Conclusions: This framework enables precise early Alzheimer’s disease (AD) diagnosis by integrating multi-scale neuroimaging features, empowering clinicians to optimize patient care through timely and targeted interventions.

Accurate and early classification of brain tumors using magnetic resonance imaging (MRI) plays a vital role in guiding timely diagnosis and therapeutic intervention. Traditional convolutional neural networks (CNNs) excel at capturing local anatomical patterns but often struggle with contextual reasoning, while Vision Transformers (ViTs) model long-range dependencies but may overlook fine-grained tumor details. To bridge this gap, we propose a hybrid deep learning framework that fuses ResNet and ViT backbones through a Cross-Gated Multi-Path Attention Fusion (CG-MAF) mechanism. CG-MAF performs dynamic, pixel-wise integration of local and global features using cross-attention and learnable gating, thereby adapting to the heterogeneous texture and spatial complexity of brain tumors. To further enhance interpretability and predictive stability, a gate-consistency loss is introduced, which enforces alignment between dual-branch attentional cues. Evaluated on a composite four-class brain MRI dataset, our model achieves state-of-the-art accuracy of 99.81% and an AUC of 0.9999. Ablation studies confirm the critical role of CG-MAF and gate-consistency loss, while Grad-CAM and Integrated Gradients provide multi-scale visual explanations that validate the model’s attention on clinically relevant tumor regions. These findings highlight the potential of our approach for robust and interpretable brain tumor classification. All source code, trained models, and experimental results are publicly available at: https://github.com/MuhammadAhmad7171/Advance-brain-tumor to support full reproducibility.