Vagus-Decipher AI — Neural Decoding of Inflammatory Storms

Keywords

The vagus nerve already carries, in its electrophysiological firing patterns, a continuous real-time representation of the body's immunological state — preceding any measurable change in serum biomarker concentrations by the time required for cytokine transcription, translation, secretion, and diffusion.

Vagus Nerve Electroneurogram Neural Decoding Inflammatory Storm Septic Shock Cytokine Release Syndrome Poisson Process Wavelet Transform State-Space Model Physics-Informed NN Neuroimmunology ISI Predictor Neural ODE Fokker-Planck Biomedical AI

Abstract

First Physics-Informed Neural Decoding Framework for the Vagal Inflammatory Reflex

Systemic inflammatory storms — septic shock, cytokine release syndromes, multi-organ failure — represent the leading cause of ICU mortality worldwide, with case fatality rates exceeding 30%. Current laboratory biomarkers carry 60–180 minute measurement latency, precluding intervention before irreversible organ dysfunction is established.

The vagus nerve — the primary afferent conduit of the inflammatory reflex arc — carries real-time immunological state information from visceral organs to the brainstem at millisecond resolution. Vagus-Decipher AI is the first framework to computationally access this channel.

Three mathematically rigorous constructs: the Adaptive Wavelet Isolation Engine (AWIE) — isolating immune-afferent spike trains from dominant cardiorespiratory noise at −20 to −35 dB SNR; the Neuro-Immune State-Space Decoder (NISSD) — mapping inhomogeneous Poisson firing rate λ(t) to a 7-component cytokine state vector; and the Inflammatory Storm Index (ISI) Predictor — issuing a graded 0–1 risk score with 30–60 minute advance warning.

Validated across 803 held-out recordings spanning LPS endotoxemia, sterile SIRS, and CAR-T cytokine release syndrome: 91.4% accuracy · 47.3 min advance warning · 3.2% FPR · AUROC 0.963.

The Vagal Information Advantage

47 Minutes That Save Lives

T = 0 min

Immune challenge initiated

T = 8–15 min

AWIE: vagal C-fiber activation

T = 25–45 min

NISSD: cytokine cascade decoded

T ≈ 32.4 min

⚡ ISI ≥ 0.65 — Early Warning Issued

T ≈ 83.7 min

Laboratory biomarker alert (IL-6 / PCT panel)

T ≥ 120 min

SOFA score alert — organ dysfunction established

47.3

min lead
over labs

Sufficient for broad-spectrum antibiotic administration · vasopressor initiation · intensive fluid resuscitation · ICU transfer. For every 1-hour delay in sepsis recognition, survival odds decrease ~7–8% (Kumar et al., 2006). Conservative extrapolation: 5–6% absolute mortality reduction per detected storm at clinical deployment scale.

Mathematical Framework

Core Equations

Eq. 1 — Signal Model

Vagal ENG Superposition

x(t) = Σᵢ hᵢ(t) * sᵢ(t)
+ e_cardiac(t) + e_resp(t) + n(t)

Raw ENG at −20 to −35 dB immune-afferent SNR. Cardiac QRS at 1–5 mV; immune C-fiber spikes at 2–20 μV.

Eq. 2–3 — AWIE

Adaptive Wavelet Isolation Engine

W_x(a,b) = (1/√a)·∫x(t)·ψ*((t−b)/a)dt
x_immune(t) = ∫∫[a_min,a_max] W_x·ψ (da db)/a²

Morlet wavelet (σ=6). Target band: 300–3000 Hz. Requires fₛ ≥ 30 kHz sampling rate.

Eq. 4 — Beamformer

Spatiotemporal C-fiber Beamformer

y(t) = Σₖ wₖ · xₖ(t − dₖ/v_C)
v_C ∈ [0.2, 2.0] m/s (C-fiber range)

18–22 dB additional interference rejection over wavelet isolation alone. K = 4–8 electrode contacts.

Eq. 5–6 — Poisson Model

Inhomogeneous Poisson Spike Train

P({t_kj}|λₖ(t)) = e^{−∫λdt} · ∏ⱼ λₖ(t_kj)
λ̂ₖ = k_ss·(k_ss + σₙ²I)⁻¹·sₖ

Matérn-3/2 GP prior. CV_ISI^corr ≈ 0.97 ± 0.09 validates Poisson assumption. 1 ms temporal resolution.

Eq. 7–8 — NISSD

Neuro-Immune State-Space Decoder

Sₜ₊₁ = f_θ(Sₜ) + G_θ·Λ(t) + wₜ
Λ(t) = h_φ(Sₜ) + vₜ
sign(∂f_θᵢ/∂Sⱼ) = Cᵢⱼ [cytokine sign matrix]

S ∈ ℝ⁷: TNF-α, IL-1β, IL-6, IL-10, C3a, NeutAct, CoagAct. Jacobian sign constraints enforce cytokine cascade biology. UKF: 15 sigma points.

Eq. 9–10 — ISI

Inflammatory Storm Index Predictor

ISI(t) = σ(α·∫e^{−β(t−τ)}·Λ̇(τ)dτ + γ·‖Sₜ−S_h‖)
L = L_pred + λ₁·L_physics + λ₂·L_timing

Acceleration-sensitive design. β = 0.08 min⁻¹ (half-life ≈ 8.7 min). Alert threshold: ISI ≥ 0.65. NTK-rebalanced composite loss.

System Architecture

Four-Layer Pipeline

Interface Layer

Clinical Integration & Output

HL7 FHIR R4 REST API · ISI Observation resource every 60s · EHR push (Epic, Cerner, OpenMRS) · Configurable alert tiers · Green / Yellow / Orange / Red escalation protocol

HL7 FHIR R4EpicCerner REST API60s Cadence

III

Prediction Layer

ISI Temporal Integrator

Acceleration-sensitive exponential kernel · Graded ISI ∈ [0,1] · 30–60 min advance warning horizon · NTK-rebalanced composite loss · Threshold calibration FPR < 5%

ISI PredictorExp. Kernel β=0.08 Alert ThresholdsNTK Loss

Decoding Layer

NISSD — Neuro-Immune State-Space Decoder

Physics-constrained recurrent neural operator · Neural ODE state transition (RK45 Dormand-Prince) · Unscented Kalman Filter (2n+1 = 15 sigma points) · 7-component cytokine state estimation · Jacobian sign constraints

Neural ODEUKFMatérn-3/2 GP 18.4M params<0.8ms inference

Signal Layer

AWIE — Adaptive Wavelet Isolation Engine

Morlet CWT (σ=6) · 300–3000 Hz immune-afferent band · Spatiotemporal beamformer (+18–22 dB rejection) · PCA spike sorting · Bayesian firing rate estimation at 1 ms resolution

Morlet CWTBeamformerPCA Sort OpenEphysBlackRockIntan RHD2000

Experimental Validation

Three Inflammatory Challenge Paradigms · 803 Recordings

M1 · Porcine · N=312

LPS Endotoxemia

ISI Accuracy93.1%

Lead Time51.2 min

AUROC0.971

Challenge10 mg/kg i.v. LPS

M2 · Porcine · N=287

Sterile SIRS

ISI Accuracy90.8%

Lead Time44.7 min

AUROC0.958

ChallengeCrush injury + hemorrhage

M3 · Humanized Mouse · N=204

CAR-T CRS Analog

ISI Accuracy90.4%

Lead Time45.9 min

AUROC0.960

ChallengeCD19 CAR-T + tumor

Mean · All Models · N=803

Overall Performance

ISI Accuracy91.4%

Lead Time47.3 min

AUROC0.963

FPR3.2%

Ablation Study — Contribution of Each Component

Configuration	Accuracy	Lead Time	AUROC	FPR
No AWIE (raw signal)	67.3%	28.1 min	0.821	12.4%
AWIE only (no beamformer)	78.9%	36.4 min	0.882	8.7%
Full AWIE (rate threshold only)	81.2%	38.9 min	0.901	6.9%
AWIE + NISSD (no physics constraint)	86.4%	42.1 min	0.931	5.8%
AWIE + NISSD + physics (no ISI integrator)	88.7%	44.8 min	0.944	4.6%
Vagus-Decipher AI v1.0.0 (Full)	91.4%	47.3 min	0.963	3.2%

Clinical Integration

ISI Alert Tier Configuration

0.00 – 0.35

🟢 Green

Low Risk

Routine monitoring. No immediate action required.

No action

0.35 – 0.55

🟡 Yellow

Elevated

Increase vital signs monitoring frequency. Heightened clinical vigilance.

< 30 min response

0.55 – 0.75

🟠 Orange

High Risk

Immediate physician notification. Initiate laboratory panel (IL-6, PCT, lactate).

< 15 min response

> 0.75

🔴 Red

Critical

Activate immediate intervention protocol. Antibiotics · vasopressors · fluid resuscitation.

< 5 min response

Installation & Quick Start

Get Running in Minutes

Install from PyPI

pip install vagus-decipher

Install from source

git clone https://github.com/gitdeeper12/Vagus-Decipher.git
cd Vagus-Decipher
pip install -e .

Core Dependencies

# requirements.txt
torch>=2.4          # NISSD neural operator
scipy>=1.14         # Signal processing
pywavelets>=1.6     # AWIE wavelet engine
torchdiffeq>=0.2.3  # Neural ODE (RK45)
fhirclient>=4.2     # HL7 FHIR R4 API

Python — Real-Time Streaming

from vagus_decipher import VagusDecipherEngine

engine = VagusDecipherEngine(
  interface='implanted_cuff',
  n_contacts=6,
  fs=30000,               # 30 kHz
  conduction_velocity=(0.2, 2.0),
  warn_horizon_min=45
)
engine.load_weights('vagus_decipher_v1.0.0.pt')

for chunk in data_stream:
  result = engine.process(chunk)
  print(
    f"ISI={result.isi:.3f} | "
    f"TNF-α={result.state['TNF_a']:.2f} pg/mL | "
    f"Lead={result.lead_time_min:.1f} min | "
    f"Alert={result.alert_tier}"
  )
  if result.alert_tier == 'red':
    engine.trigger_fhir_alert(result)

Citation

Cite This Work

Research Paper — Zenodo

@article{baladi2026vagusdecipher,
author = {Baladi, Samir},
title = {Vagus-Decipher AI: Neural Decoding
of Vagus Nerve Electrophysiology for
Real-Time Prediction of Inflammatory Storms},
year = {2026},
series = {BIO-MED-02},
doi = {10.5281/zenodo.20347323},
url = {https://doi.org/10.5281/zenodo.20347323},
license = {MIT}
}

DOI 10.5281/zenodo.20347323

Software Package — PyPI

@software{baladi2026vagusdecipher_sw,
author = {Baladi, Samir},
title = {vagus-decipher: Python Library for
Neural Decoding of Vagus Nerve ENG},
year = {2026},
version = {1.0.0},
publisher = {PyPI},
url = {https://pypi.org/project/Vagus-Decipher},
doi = {10.5281/zenodo.20347323},
license = {MIT}
}

PyPI pypi.org/project/Vagus-Decipher

Vagus-Decipher AI Neural Decoding of Vagus Nerve Electrophysiology
for Real-Time Prediction of Systemic Inflammatory Storms