Open AccessA Hybrid Deep Learning Framework For Early-stage Alzheimer’s Disease Classification From Neuro-imaging Biomarkers
Author(s) -
Samina Akram,
Muhammad Amjad Iqbal,
Muhammad Rashid,
Muhammad Shahid Bhatti,
Benish Fida
Publication year - 2025
Publication title -
ieee access
Language(s) - English
Resource type - Magazines
SCImago Journal Rank - 0.587
H-Index - 127
eISSN - 2169-3536
DOI - 10.1109/access.2025.3574039
Subject(s) - aerospace , bioengineering , communication, networking and broadcast technologies , components, circuits, devices and systems , computing and processing , engineered materials, dielectrics and plasmas , engineering profession , fields, waves and electromagnetics , general topics for engineers , geoscience , nuclear engineering , photonics and electrooptics , power, energy and industry applications , robotics and control systems , signal processing and analysis , transportation
Alzheimer’s disease, a leading neurodegenerative disorder, poses significant challenges to modern healthcare, emphasizing the urgent need for precise early-stage diagnostics. Despite promising insights from neuroimaging into potential AD biomarkers, the intricacy of data interpretation demands advanced computational techniques. Existing diagnostic tools have yet to fully harness the combined strengths of neuroimaging and modern deep-learning architectures for nuanced, early-stage AD classification. In this study, we propose a novel hybrid deep learning framework that ensembles three state-of-the-art architectures—EfficientNetB7, Xception, and MobileNetV3Large—through a soft-voting strategy to classify Alzheimer’s disease across multiple stages. Our ensemble methodology deciphers complex patterns within Magnetic Resonance Imaging, mapping them to specific AD early stages. Integrating pre-trained model weights and adaptive learning rates ensures swift convergence and optimal performance. Benchmarking against a comprehensive AD neuro-imaging dataset revealed that our ensemble approach achieves a groundbreaking precision of 99.87%. These results significantly surpass those of individual architectures addressing this problem. Evaluation metrics like precision-recall and f1 score nearly approached 100% across different experiment stages. With inference times as low as 3ms for MobileNetV3Large, our model is highly suitable for real-time deployment in clinical environments. Furthermore, Grad-CAM explainability is incorporated to highlight critical brain regions contributing to the model’s predictions, improving clinical trust. To the best of our knowledge, this is the first ensemble-based deep learning framework combining EfficientNetB7, Xception, and MobileNetV3Large for multi-stage Alzheimer’s classification. The proposed hybrid model can confidently replace conventional diagnostic methods. As medical science is quickly changing and adapting AI decisions, our methodology is helpful for early diagnosis of dementia detection in the early stages of preemptive treatment.