| name | hcq-alzheimer-classification-vae-quantum-kernels |
| description | Hybrid Classical-Quantum pipeline for Alzheimer's classification using supervised β-VAE and quantum kernels (arXiv:2606.14194) |
| category | quantum-medical |
Hybrid Classical-Quantum Alzheimer's Classification
Methodology from arXiv:2606.14194 (June 2026). Two-stage HCQ pipeline for binary AD classification from 3D structural MRI volumes.
Core Pattern
Classical and quantum components designed to complement rather than operate independently:
- Supervised 3D β-VAE compresses MRI volumes into latent code using voxel-wise reconstruction + KL-divergence + focal classification losses
- PLS regression selects disease-separating components and rescales into rotation angles
- ZZ quantum feature map encodes onto qubit register
- Precomputed-kernel SVM on quantum Gram matrix for classification
Key Findings
- 72.1% accuracy, 0.799 AUC on 308 ADNI-1 subjects (AD vs CN)
- Cross-fold variance halved with stability-enhanced variant
- Novelty: quantum kernel operates on disease-aware features learned end-to-end by supervised autoencoder
- 3D Grad-CAM validates model focus on Alzheimer's-linked brain regions
Implementation Steps
- Resize 3D MRI volumes to standard dimensions (96×96×96)
- Train supervised 3D β-VAE with reconstruction + KL + focal losses
- Extract 64-dimensional latent codes
- Apply PLS to select 6 components best separating classes
- Encode components as rotation angles via ZZ quantum feature map
- Compute N×N Gram matrix from quantum state overlaps
- Train SVM on Gram matrix for classification
- Validate with 3D Grad-CAM for biological plausibility
When to Use
- Binary classification from 3D medical imaging
- Scenarios where classical methods struggle with high-dimensional data
- Need for interpretable, biologically-grounded quantum ML
- Diagnostic classification across biomedical imaging domains
References
- arXiv: 2606.14194v1
- Authors: Tia Tiwari, Vamshi Krishna Kancharla, Neelam Sinha
