.png){kind=icon}
.png){kind=icon}
Return
HyperTransFusion: a hypernetwork transformer with black winged kite optimization for multimodal early Alzheimer's disease diagnosis
.png){kind=icon}
OA
AI
S
P
DOI：10.3389/fdgth.2026.1874626.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
IntroductionEarly detection of Alzheimer's disease (AD) is critical for timely intervention and effective disease management. Existing diagnostic systems are often limited by unimodal data or static fusion strategies that fail to capture complex interactions between neuroimaging and cognitive biomarkers.MethodsA HyperTransFusion framework was proposed for AD classification using MRI scans and cognitive assessment scores from the ADNI dataset. The framework integrates a Hypernetwork-based Transformer for adaptive multimodal fusion; a Vision Transformer (ViT) for MRI feature extraction; dense embedding layers for cognitive features; and a BWKO-SA optimization strategy combining Black-Winged Kite Optimization and Simulated Annealing for hyperparameter tuning. Performance was evaluated using five-fold patient-wise cross-validation.ResultsThe proposed framework achieved a macro AUC of 0.9311 ± 0.039 and an overall accuracy of 0.8624 (95% CI: 0.7691-0.9558); demonstrating robust generalization across unseen subjects. Ablation studies confirmed the contribution of the Hypernetwork and BWKO-SA modules. Slice-level evaluation achieved a peak accuracy of 99.1%; reported only as a supplementary indicator of feature separability.DiscussionThe proposed HyperTransFusion framework enables adaptive multimodal integration and improves the accuracy of early AD classification. Further validation using independent multi-center cohorts is required before clinical translation.
Keywords:
vision transformer
Alzheimer disease
multimodal fusion
disease detection
adaptive fusion
Journal
F
IF:
3.8
Papers:
2.0K
Citations:
3.1K
Organization
No organization information available
Cited Papers