A Hybrid M-NET And U-NET Framework With Tissue-Aware Feature Fusion For Alzheimer’s Disease Prediction From Brain MRI

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that poses a significant global healthcare challenge, particularly among the aging population. Accurate and early diagnosis remains difficult due to subtle structural changes in brain tissue and the limitations of conventional clinical assessments. To address these challenges, this paper presents a hybrid deep learning framework for Alzheimer’s disease prediction that integrates tissue-aware brain MRI segmentation, texture feature extraction, and feature fusion-based classification within a unified pipeline.Initially, brain MRI images are preprocessed using contrast enhancement and intensity normalization to improve tissue visibility and learning stability. An M-Net architecture enhanced with Deep Embedded Clustering (DEC) is then employed to achieve accurate segmentation of White Matter (WM), Gray Matter (GM), and Cerebrospinal Fluid (CSF), enabling reliable isolation of disease-relevant brain regions. From the segmented WM and GM tissues, Gray Level Co-occurrence Matrix (GLCM)–based texture features are extracted to characterize microstructural and intensity variations associated with neurodegeneration.To effectively integrate complementary information from multiple brain tissues, a U-Net-based feature fusion network is introduced to combine WM and GM features while preserving spatial coherence. The fused feature representation is subsequently used for multi-class classification of Alzheimer’s disease, Mild Cognitive Impairment (MCI), and Healthy Control (HC) subjects. Experimental evaluation on the ADNI dataset demonstrates that the proposed framework achieves an accuracy of 94.18%, along with high precision, recall, F1-score, and an AUC of 0.94, outperforming conventional segmentation and classification baselines.The results highlight that accurate tissue segmentation combined with texture-aware feature fusion significantly enhances Alzheimer’s disease prediction, particularly for early-stage diagnosis. The proposed framework provides a reliable and interpretable computer-aided diagnostic solution with strong potential for clinical decision support