Cross-Scale Feature Learning for Alzheimer's Disease Detection via Hybrid Deep Networks

Authors

• Essam Haidar Majeed, Department of Computer Systems Techniques, Technical Institute of Najaf, Al-Furat Al-Awsat Technical University, Najaf, 54001, IraqFollow
• Fallah H. Najjar, Department of Computer Systems Techniques, Technical Institute of Najaf, Al-Furat Al-Awsat Technical University, Najaf, 54001, Iraq AND Department of Emergent Computing, Faculty of Computing, Universiti Teknologi Malaysia, 81310 UTM Johor Bahru, Johor, MalaysiaFollow
• Safaa M. Kareem, Department of Computer Systems Techniques, Technical Institute of Najaf, Al-Furat Al-Awsat Technical University, Najaf, 54001, IraqFollow
• Munaf H. Kareem, Department of Computer Systems Techniques, Technical Institute of Najaf, Al-Furat Al-Awsat Technical University, Najaf, 54001, IraqFollow
• Duha Amer Mahdi, Department of Computer Systems Techniques, Technical Institute of Najaf, Al-Furat Al-Awsat Technical University, Najaf, 54001, IraqFollow

Abstract

Alzheimer's disease (AD) is a chronic neurodegenerative disorder that is the leading causeof dementia. Early diagnosis is essential for interepting the progressionof cognitive decline and favoring the clinical outcome, although conventional diagnostic means may not be able to identify early symptomatic forms. In this manuscript, we propose a residual attention-based hybrid deep learning model withthe EfficientNetB1 backbone for improving the classification of AD based on structural MRI data obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and ADNI1 datasets. The network structure integrates two-dimensional (2D) convolutional feature extraction, multi-head self-attention for building the global context information, and residual learning. The proposed method can acquire both the localanatomical characters and the global brain structure dependencies effectively. It outperforms several state-of-the-art methods in terms of diagnostic accuracy on both models, reaching macro-averaged accuracy of 99.13% and 99.37% over ADNI and ADNI1 data, respectively, based on three anatomical planes (axial,sagittal, and coronal). This then led to biologically feasible areas from attention maps, whichenhanced interpretability. These findings indicate that our model is an effective and interpretable approach to early and accurate classification of AD, offering the potential to be deployed inclinical settings. In future, the model will be evaluatedfor its usability and pragmatic effectiveness in clinical use to confirm sufficient applicability for neurologists and radiologists.

Keywords

Alzheimer's disease, Cross-scale feature, Classification, Deep learning, Hybrid model

Subject Area

Computer Science

Article Type

Article

First Page

1779

Last Page

1794

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

How to Cite this Article

Majeed, Essam Haidar; Najjar, Fallah H.; Kareem, Safaa M.; Kareem, Munaf H.; and Mahdi, Duha Amer (2026) "Cross-Scale Feature Learning for Alzheimer's Disease Detection via Hybrid Deep Networks," Baghdad Science Journal: Vol. 23: Iss. 5, Article 16.
DOI: https://doi.org/10.21123/2411-7986.5300