Fetal Twin

A ground-truth-labeled, mechanistic in-silico testbed for fetal physiological development โ€” enabling biomarker discovery where clinical data cannot go.

Open Testbed Antepartum + Intrapartum Ground-Truth Labeled Noncommercial License
๐Ÿ”ฌ

Mechanistic, not generative

Physiology is simulated from first principles โ€” cardiovascular, metabolic, autonomic โ€” not learned from data.

๐Ÿท๏ธ

Latent ground truth

pH, lactate, perfusion pressure, and injury states are co-registered with every synthetic monitoring signal.

๐Ÿงช

In-silico controls

Disable noise, ablate reflexes, quantize signals โ€” interventions impossible in live subjects.

๐Ÿ“

Instrument, not a digital twin

Calibration-grade, population-level. Not a validated patient-specific clinical predictor.

The structural gap: Ground-truth mechanistic labels cannot be ethically collected at scale during human pregnancy or labor. Fetal Twin bypasses this by providing a controlled environment where every latent variable is known.
โš ๏ธ

Detector artifacts can fabricate HRV

Naive beat detectors inject spurious beats scaling as tยทแนช/Tยฒ under a time-varying period. Up to 94.5% of detected beats can be spurious โ€” making apparent HRV amplitude a property of the detector, not the fetus.

๐Ÿ”

HRV requires an explicit stochastic driver

Beat-to-beat variability does not arise as a self-sustained oscillation of the deterministic autonomic loop โ€” it requires an injected noise source.

๐Ÿ“ก

Fidelity-matched biomarker selection

Deceleration area survives 4 Hz CTG sampling. RMSSD is corrupted by quantization jitter at 4 Hz and requires fetal-ECG-grade acquisition.

โฑ๏ธ

Early warning via autonomic exhaustion

HRV-based predictors track autonomic exhaustion ahead of metabolic decompensation โ€” providing a regime-dependent lead time before pH threshold crossing.

๐Ÿซ€

Rise-then-collapse motivates a pacemaker hypothesis

Autonomic modulation alone cannot reproduce the published RMSSD rise-then-collapse trajectory โ€” motivating, but not proving, an intrinsic sinoatrial pacemaker contribution.

Fetal Twin integrates five physiological layers, each contributing to the synthetic monitoring signal.

Cardiovascular

Heart rate, blood pressure, umbilical cord occlusion response

Mechanistic

Metabolic / Acid-Base

pH, lactate, aerobic/anaerobic metabolism, buffer dynamics

Mechanistic

Autonomic

Fast/slow vagal split, sympathetic drive, chemoreflex, exhaustion

Mechanistic

HRV / Pacemaker

AR(2) intrinsic surrogate โ€” coupled-clock sinoatrial model in development

Surrogate

Gestational Development

Antepartum scaling, growth, organ maturation across gestation

Mechanistic

33-state ODE system  ยท  Explicit integer seeds  ยท  Reproducible to numerical tolerance  ยท  Phase-accumulator beat detector

ODE state variables

33

Spurious beats (naive detector)

94.5%

CTG sampling (RMSSD fails)

4 Hz

DTH lead time (moderate severity)

~57 min

Synthetic signal stack

FHR traceโœ“ robust at 4 Hz
Deceleration areaโœ“ robust at 4 Hz
RMSSDโš  requires fECG-grade
pH / lactate (latent)โˆž always available
Source code ยท GitHub โ†— Archived ยท Zenodo DOI โ†—