Virtual Human Twins (VHTs) for integrated clinical decision support in prevention and diagnosis HORIZON-HLTH-2027-03-TOOL-04

NeuroVHT: Multi-scale Virtual Human Twins for early screening and personalised risk stratification in Alzheimer’s and Parkinson’s disease

Vision. Neurodegenerative diseases (Alzheimer’s disease—AD; Parkinson’s disease—PD) are long, silent trajectories where prevention and early diagnosis are bottlenecked by fragmented evidence (single test, single modality, single visit). We propose NeuroVHT, a multi-scale, longitudinal Virtual Human Twin (VHT) that integrates brain structure/function, electrophysiology, and everyday digital biomarkers to deliver clinically meaningful decision support for early screening, differential risk stratification, and referral prioritisation—particularly for populations with diverse characteristics and care access.

Objectives (what we will deliver)

O1 — Patient-specific NeuroVHT models that are multi-scale (behaviour → signals → imaging) and longitudinal (updates with new evidence over time), producing an interpretable risk state and confidence.
O2 — Integrated multi-modal screening pipeline that fuses digital biomarkers (pen/keyboard/gait/voice), EEG, and multimodal neuroimaging (MRI+PET) into a unified VHT representation.
O3 — Clinically validated decision support for prevention/diagnosis: earlier identification of high-risk individuals, reduced false referrals, and support for personalised diagnostic workups.
O4 — Interoperable assets made available to the European VHT ecosystem (models, pipelines, documentation, evaluation protocols), aligned with platform specifications and community practices.

Approach

B1. Multi-modal data backbone. Combine: (i) neuroimaging (sMRI, FDG-PET), (ii) electrophysiology (resting-state EEG; visual stimulus EEG), (iii) digital biomarkers from at-home or low-cost tests: handwriting spirals, keystroke dynamics, gait accelerometers, and voice.
B2. Representation-first modelling. Build on our demonstrated advantage of transforming raw signals into information-rich representations (e.g., time–frequency images) and then applying deep models/transfer learning for robust screening.
B3. Hybrid VHT modelling.