NERVALYS-AI is a computational precision stratification platform that reads the hidden biological structure of diabetic peripheral neuropathy — per patient, per mechanism, per visit.
Every output derives from a single Personalized Vector — a patient-specific latent representation of how disease behaves biologically.
This patient's PV encodes high-complexity biological state with asymmetric distribution — indicating multi-mechanism involvement and active disease dynamics.
Rapid progression · Inflammatory · Hyper mismatch
| Distal | Proximal | Global | Reinn. | |
|---|---|---|---|---|
| Inflammatory | 0.76 | 0.42 | 0.55 | 0.31 |
| Axonal loss | 0.81 | 0.61 | 0.89 | 0.28 |
| Compensatory | 0.29 | 0.71 | 0.46 | 0.78 |
| Metabolic | 0.64 | 0.58 | 0.69 | 0.34 |
Four engineered stages convert heterogeneous multimodal patient data into a structured, interpretable disease representation.
Inflammatory biomarkers, oxidative stress markers, EMG signals, and clinical metadata are normalized and validated.
Biological signals pass through a denoising Autoencoder. EMG through 1D-CNN + BiLSTM. Each modality encoded separately to preserve causal structure.
Cross-modal Transformer attention fuses latents into the Personalized Vector — a structured disease representation unique to each patient.
Multi-head outputs deliver stratification axes, prognosis, risk windows, nerve age, and GNN-based mechanism-to-muscle interaction maps.
Validated on an independent human cohort of 90 patients — zero retraining. Structure generalizes because the biology generalizes.
AUROC drop from internal to external: only −0.022. Structural generalization preserved across independent human cohort with zero retraining.
Macro-F1 of 0.954. Sensitivity 94.8%, specificity 96.7%. Balanced performance across stratification profiles — no class collapse.
Patient-level stratification profile stability across repeated inference: 91% consistency. Biological interpretation remains robust.
Mechanism-aware stratification replaces binary classification — clinicians now see *how* and *why* disease progresses per patient.
39% of patients reclassified into actionable stratification profiles — moving beyond symptomatic/asymptomatic into mechanism-matched intervention.
Earlier, mechanism-matched intervention reduces cumulative nerve damage — the root cause of foot ulceration and amputation.
Avoiding unnecessary escalation across a 130M-person disease population translates to hundreds of millions in avoided expenditure globally.
Access the full research, source code, and app demonstration below.
NERVALYS-AI is seeking clinical pilot partners, institutional collaborators, and strategic co-investors to bring mechanism-aware DPN stratification into real-world care.