Related Experiment Video
Updated: Aug 28, 2026
DeepOmicsAE: Representing Signaling Modules in Alzheimer's Disease with Deep Learning Analysis of Proteomics, Metabolomics, and Clinical Data
Published on: December 15, 2023
HyperTransFusion: a hypernetwork transformer with black winged kite optimization for multimodal early Alzheimer's
S Sabari Vasan1, P Jayalakshmi1
1School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, India.
Introduction:
Early 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.
Methods:
A 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.
Results:
The 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.
Discussion:
The 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.