Methodology

The science behind
cardiac glycemic sensing.

A systematic investigation spanning ICU databases, wearable cohorts, and 51-dimensional feature spaces. Peer-reviewed, reproducible, and designed for clinical deployment.

Foundations

Four principles. One pipeline.

The physiological, engineering, and clinical premises that make ECG-based glucose estimation theoretically tractable and practically useful.

01Physiology

Cardiac-Glycemic Coupling

Blood glucose concentration directly modulates cardiac electrophysiology. Hyperglycemia prolongs QTc intervals through potassium channel disruption; hypoglycemia activates sympathetic responses measurable in HRV. These effects are weak, overlapping, and confounded — but they are consistent enough to learn from at scale.

02Signal Processing

Temporal Alignment Problem

ECG-CGM pairing is not trivial. CGM readings reflect interstitial glucose with a 10–20 minute physiological lag behind blood glucose. ECG morphology reflects instantaneous cardiac state. Correctly aligning these two timeseries requires careful window selection, lag correction, and validation against simultaneous blood draws.

03Validation

Clinical-Scale Validation

Small wearable datasets (D1NAMO: 9 subjects) reveal feasibility. ICU databases (MIMIC-IV: 131K+ patients) reveal robustness under pharmacological confounding, metabolic extremes, and device variability. We validate on both — and report Clarke Error Grid Zone distributions, not just R² or MAE.

04Clinical AI

Safety-Aware Loss Functions

Standard regression losses (MSE, MAE) are clinically naive — they penalize a 50 mg/dL error equally at euglycemia and at a critical hypoglycemic threshold. SAGE-Net's loss function engineering weights errors by their clinical zone, trading overall RMSE for dramatically improved sensitivity in dangerous ranges.

Feature Engineering

51 dimensions. One signal.

Multi-domain feature extraction transforms raw ECG into a structured representation across HRV, morphology, and statistical domains.

HRV Time Domain

  • ◆SDNN
  • ◆RMSSD
  • ◆pNN50
  • ◆Mean RR
  • ◆RR Triangular Index

HRV Frequency Domain

  • ◆LF Power
  • ◆HF Power
  • ◆LF/HF Ratio
  • ◆VLF Power
  • ◆Total Power

ECG Morphology

  • ◆PR Interval
  • ◆QRS Duration
  • ◆QT/QTc
  • ◆T-wave Area
  • ◆ST Deviation

Statistical

  • ◆Skewness
  • ◆Kurtosis
  • ◆Signal Entropy
  • ◆Fractal Dimension
  • ◆Autocorrelation
Validation

Tested at scale, not just in theory.

Validation across multiple independent datasets with distinct populations, devices, and clinical contexts.

Dataset

MIMIC-IV + MIMIC-IV-ECG

Scope

131K+ patients, ICU multi-site

Approach

BigQuery linkage, 15-step QC pipeline

Outputs

Pharmacological confounding strata, signal quality tiers

Dataset

AI-READI v3.0.0

Scope

1,552 T2D participants

Approach

Philips PageWriter TC30 ECG + Dexcom G6 CGM pairing

Outputs

Cross-dataset zero-shot eval on OhioT1DM

Dataset

D1NAMO

Scope

9 subjects, type 1 diabetes

Approach

Temporal alignment optimization study

Outputs

Lag window sweep, segment-level glucose correlation

Dataset

eICU Collaborative Research Database

Scope

200K+ ICU admissions, multi-hospital

Approach

External validation in SAGE-Net pipeline

Outputs

Clarke EGA Zone A/B classification across 48 model configs

Read the full research.

Peer-reviewed papers across IEEE, ACM, and Elsevier.