DUOFE NET:Diagnosis of Alzheimer Disease Detection via Deep Learning based Duo Feature Extraction Model

Alzheimer's disease (AD) causes progressive neurodegeneration and cognitive decline among the elderly, making early detection crucial for effective treatment. However, existing AD classification models often struggle to capture subtle structural variations and rely on single-view feature extraction, resulting in reduced robustness and diagnostic accuracy. To address these limitations, a novel DuoFE Net is proposed for precise multi-stage AD classification by integrating both local and global brain structural information from dual-view MRI scans. The model employs an Adaptive Bilateral Histogram Equalization (ABIHE) filter to enhance MRI quality by suppressing noise while preserving anatomical edges. A Refinement Network with Global-Local Attention (GLA) is introduced to extract discriminative foreground-background features by modelling contextual dependencies and fine-grained tissue variations. LinkNet utilizes the refined dual-view features to accurately classify NC, MCI, and AD stages by preserving crucial spatial information through its efficient encoder-decoder architecture. The proposed DuoFE Net method has achieved an accuracy rate of 99.38% for AD detection. The proposed LinkNetenhances the total accuracy by 4.42%, 2.21%, 3.42%, 2.06%, 1.52%, 1.12%, 1.02%, and 0.80% better than DenseNet, AlexNet, MobileNet, ShuffleNet, ConvNeXt, CoAtNet, Vision Transformer and Swin Transformer respectively. The proposed DuoFE Net improvestotal accuracy by 2.56%, 5.40%, 22.69%, 1.42% and 10.97% compared to Xception, Explainable AI, XGBoost, ViT and LightGBM respectively.