Model

ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime)

UnconfirmedCatalog paper date: 2026Added to atlas: Nov 19, 2026

Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.

Single-lead ECG stream model (250 Hz, 10 s windows): per-window AF "detect" probability and onset probability within 1, 2, 5, 10, 30 min. Neural single- and two-lead fold-0 bundles plus a LightGBM baseline; alarm = causal mean of last 3 windows over thresholds set for 1 and 0.2 false alarms/h. Live in-browser demo Space.

Parameters
0.92M
Signals
ECG
License
CC BY-NC-SA 4.0 (weights; because Icentia11k is NC-SA)
Year
2026
Architecture

pretrained Transformer window encoder (0.8M params, 0.2 s patches, lead embedding) + GRU (0.12M) over successive 10 s windows; LightGBM baseline on 49 features

At a glance
Type
Task model
Modalities
ECG
Form factor
Not reported
Parameters
0.92M
Architecture
pretrained Transformer window encoder (0.8M params, 0.2 s patches, lead embedding) + GRU (0.12M) over successive 10 s windows; LightGBM baseline on 49 features
Venue
Not reported
Year
2026
arXiv
Not reported
DOI
Not reported
Code
Not released
Licence
CC BY-NC-SA 4.0 (weights; because Icentia11k is NC-SA)
Caveats
Not reported
Added
Nov 19, 2026
Updated
Nov 19, 2026

Original description

Small models that read a single-lead ECG as a stream of 10-second windows (250 Hz, 2500 samples, millivolts) and give, for every window, the probability that atrial fibrillation (AF) is under way ("detect") and that it will start within 1, 2, 5, 10 or 30 minutes.

Research models, not a medical device. Do not use them to make decisions about anyone's care.

Verbatim fromHugging Face· Description from the Hugging Face card; licence as stated on the card

ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime): ChatPPG research summary card

How it works

How ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime) works: Input signals → Model → Downstream tasksInput signals: ECG (Electrocardiogram). Datasets: Long-Term AF Database (LTAFDB, PhysioNet), SHDB-AF, Icentia11k single-lead continuous raw ECG (PhysioNet), IRIDIA-AF, CPSC 2021 Paroxysmal AF Database (PhysioNet). Model: ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime), 0.92M parameters (Transformer · Gradient-boosted trees). Downstream tasks: AF / rhythm.Input signalsECG: ElectrocardiogramECGElectrocardiogramDatasets: Long-Term AF Database (LTAFDB, PhysioNet), SHDB-AF, Icentia11k single-lead continuous raw ECG (PhysioNet), IRIDIA-AF, CPSC 2021 Paroxysmal AF Database (PhysioNet)DatasetsLong-Term AF Database (LTAFDB, PhysioNet)…Model×Necg-af-early-warning…0.92M parametersTransformer · Gradient-boosted treesDownstream tasksAF / rhythmAF / rhythm

Datasets linked in the hub; pretraining vs evaluation roles follow the paper.

Architecture
pretrained Transformer window encoder (0.8M params, 0.2 s patches, lead embedding) + GRU (0.12M) over successive 10 s windows; LightGBM baseline on 49 features
Inputs
ECG
Approach
ecg · afib · early warning · streaming · wasm

Datasets used· 4

Cite this model

Cite the publication recorded for this model. Check it against the publisher before submitting.

ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime). (2026).
@misc{ecgafearlywarnin2026ecgafearlywarning,
  title         = {{ecg-af-early-warning (single-lead streaming ECG AF detection + 1–30 min onset early warning; Transformer window encoder + GRU, LightGBM baseline, Rust/WASM runtime)}},
  year          = {2026}
}

Source metadata: 2026

Record updated Nov 19, 2026

History

  1. Added to the ChatPPG research hub