Gradual Variation-Based Dual-Stream Deep Learning for Spatial Feature Enhancement With Dimensionality Reduction in Early Alzheimer’s Disease Detection
Alzheimer’s disease (AD) is one of the most common neurological brain disorders, developing gradually and primarily causing the deterioration of brain function, leading to memory loss and behavioral changes. Early detection through computer-aided systems is crucial for effective intervent...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10890952/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Alzheimer’s disease (AD) is one of the most common neurological brain disorders, developing gradually and primarily causing the deterioration of brain function, leading to memory loss and behavioral changes. Early detection through computer-aided systems is crucial for effective intervention and treatment, providing automated and highly accurate diagnoses of Alzheimer’s by analyzing brain scans, such as MRI, to assess atrophy in key regions like the hippocampus and entorhinal cortex. Recently, many researchers have been developing computer-aided AD recognition systems using various technologies. However, existing approaches face challenges in achieving good performance due to difficulties in capturing complex spatial features and handling high-dimensional data. This study proposes a Gradual Variation-Based Dual-Stream Deep Learning (DL) framework for Spatial Feature Enhancement with Dimensionality Reduction to address these challenges. The method begins by converting MRI voxel data into slice images, followed by the extraction of gradual spatial variations based on the Haar Wavelet Transform (WT) to capture large-scale structural changes. These features are processed through a dual-stream DL model, where each stream captures complementary spatial features. The concatenated features are refined using a channel attention (CA) module, which selects the most discriminative features while performing dimensionality reduction to retain only the most important information, eliminating redundant or less informative features. Finally, a DL-based classification module is applied to detect early signs of Alzheimer’s disease. We evaluated the model on two datasets, Complete 1Yr 1.5T and Complete 3Yr 3T, from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), using MRI scans from three planes (axial, sagittal, and coronal). The proposed method achieved 99.05% accuracy, 99.05% precision, 99.05% recall, 99.73% ROC-AUC, and 99.33% F1-score, significantly outperforming state-of-the-art methods. These results demonstrate the robustness and computational efficiency of our approach for early Alzheimer’s diagnosis, offering a valuable tool for clinical applications. |
|---|---|
| ISSN: | 2169-3536 |