ECGDetect-AI 2026

Big challenge 02 · AI in Health

Beyond Rhythm Recognition: Cardiac Abnormality Detection and ECG Interpretation from 12-Lead Signals

Team registration openSubmissions open

ECG

Category: Artificial Intelligence in Healthcare
Format: Online, November – December 2026 · final presentations on 17 December
Tracks: 2 ranked + demo · up to 3 per team
Teams: Solo or teams of up to 4
Submission deadline: 10 December 2026

Overview

Electrocardiography (ECG) is a widely used examination for assessing the electrical activity of the heart. A standard 12-lead ECG can reveal changes associated with myocardial infarction, conduction disturbances, hypertrophy and cardiac rhythm disorders.

Automated ECG interpretation remains difficult because abnormalities may occur together, recordings differ in signal quality, and some conditions are much less frequent than others. A model must identify relevant patterns across several leads rather than assign one diagnosis to every recording.

ECGDetect-AI 2026 invites participants to develop methods for analysing clinical ECG signals and producing consistent predictions for diagnostic and rhythm labels.

The competition uses PTB-XL, a publicly released dataset of 21,799 ECG recordings from 18,869 patients.

Objectives

01Develop accurate methods for automatic analysis of 12-lead ECG recordings.
02Identify diagnostic abnormalities using multi-label classification.
03Recognise cardiac rhythms from ECG waveforms.
04Handle recordings with multiple findings and imbalanced label frequencies.
05Compare signal-processing methods and deep learning architectures.
06Explore models suitable for practical and efficient inference.
07Encourage reproducible research in AI-assisted electrocardiography.

The data

PTB-XL, a Large Publicly Available Electrocardiography Dataset — Scientific Data, 2020. The competition uses the corrected PhysioNet release, version 1.0.3 (2022).

Dataset source

Signal modality: 12-lead electrocardiography (ECG)
Recordings: 21,799
Patients: 18,869
Recording duration: 10 seconds
Sampling frequencies: 100 Hz and 500 Hz
ECG annotations: Diagnostic, rhythm and form statements
Data format: WFDB waveform files and CSV metadata
Suggested partitions: Folds 1–8: training · fold 9: validation · fold 10: test
Licence: Creative Commons Attribution 4.0 (CC BY 4.0)

PTB-XL contains patient identifiers and recommended patient-separated folds. Its annotations use SCP-ECG statements, which allow several labels to be assigned to the same recording.
Dataset use and redistribution must preserve attribution, the licence notice and information about any modifications.
The published test labels are also public. The competition therefore uses an audited benchmark protocol rather than describing the published test fold as a hidden test set. A blind leaderboard would require an additional independently annotated dataset.

Tracks

A team may enter up to 3 of these. Each ranked track has its own leaderboard.

01Track A

Leaderboard · primary metric: macro AUROC

Multi-Label Cardiac Abnormality Detection

Develop a model that takes a 10-second, 12-lead ECG recording as input and predicts the presence of diagnostic findings across the five diagnostic superclasses defined in PTB-XL.

Labels

NORMNormal ECG
MIMyocardial infarction
STTCST/T changes
CDConduction disturbance
HYPHypertrophy

Because more than one diagnostic superclass may be associated with a recording, the output must contain five independent probabilities.

Main technical challenges

Distinguishing subtle morphological changes in ECG waveforms.
Learning from unequal class frequencies.
Handling simultaneous diagnostic statements.
Maintaining consistent predictions across the 12 leads.
Reducing sensitivity to noise and recording artefacts.

Evaluation

Metric	Purpose
Macro AUROCPrimary	Measures discrimination across the five diagnostic superclasses; primary ranking metric.
Macro Average Precision (AP)	Assesses precision-recall performance, especially for less frequent labels.
Macro F1-score	Measures the balance between precision and recall after validation-based threshold selection.
Per-class sensitivity	Measures the proportion of positive recordings recovered for each diagnostic class.

Diagnostic probabilities are evaluated independently for each class. Ties are resolved using the secondary metric, then inference time under the official evaluation environment.

Expected output

A prediction file containing five diagnostic probabilities for each ECG recording.

You submit

Track A predictions
One row per requested ECG, probabilities between 0 and 1.
ecg_id,p_NORM,p_MI,p_STTC,p_CD,p_HYP
CSV · up to 5 MB
Trained model weights & dependencies
Over 10 MB? Share a download link (Drive, Hugging Face, Zenodo…).
PT, PTH, ONNX, H5, KERAS, TFLITE, ZIP · up to 10 MB, or a link

02Track B

Leaderboard · primary metric: macro F1

Cardiac Rhythm Recognition

Develop a method that identifies cardiac rhythm statements from the same 12-lead signals.

Labels

SRSinus rhythm
AFIBAtrial fibrillation
STACHSinus tachycardia
SBRADSinus bradycardia
SARRHSinus arrhythmia

These labels are defined in the PTB-XL statement dictionary. The organisers will verify the number of positive recordings in each partition and confirm the final label set before the competition begins.

Main technical challenges

Identifying differences in rhythm and heart rate.
Detecting less frequent rhythm labels.
Managing noise and baseline drift.
Distinguishing normal rhythm from rhythm alterations.
Processing signals consistently across different sampling frequencies.

Evaluation

Metric	Purpose
Macro F1-scorePrimary	Measures average positive-class F1 across the selected rhythm labels; primary ranking metric.
Macro AUROC	Evaluates the discrimination of rhythm probabilities.
Per-class recall	Measures the detection rate for each rhythm statement.
False-positive rate	Measures incorrect positive rhythm predictions.

Decision thresholds for F1 are selected on the validation set and fixed before test evaluation. The organisers publish the label-mapping script and metric implementation before submissions open.

Expected output

A prediction file containing a probability for each specified rhythm statement.

You submit

Track B predictions
One row per requested ECG, probabilities between 0 and 1.
ecg_id,p_SR,p_AFIB,p_STACH,p_SBRAD,p_SARRH
CSV · up to 5 MB
Trained model weights & dependencies
Over 10 MB? Share a download link (Drive, Hugging Face, Zenodo…).
PT, PTH, ONNX, H5, KERAS, TFLITE, ZIP · up to 10 MB, or a link

+Optional

Advanced demonstration · not a leaderboard

Complete ECG Analysis

Teams may combine both tasks in one model and present an interface showing the ECG waveform, predicted diagnostic labels and rhythm findings. This is an advanced demonstration category rather than a mandatory leaderboard.

You submit

Demonstration (video or live link)optional
A short screen recording or a link to the running interface.
Link

How to submit

Register your team, then upload each deliverable from the participant portal. Prediction files are checked the moment you upload them, and your latest upload before the deadline is the one evaluated.

Suggested approaches

Signal-processing features with classical classifiers
One-dimensional convolutional neural networks (1D CNNs)
Residual and Inception-style time-series networks
Transformer-based ECG architectures
Recurrent or state-space sequence models
Hybrid signal-processing and deep learning approaches

Track A · Multi-Label Cardiac Abnormality Detection

Track A predictions
One row per requested ECG, probabilities between 0 and 1.
ecg_id,p_NORM,p_MI,p_STTC,p_CD,p_HYP
CSV · up to 5 MB
Trained model weights & dependencies
Over 10 MB? Share a download link (Drive, Hugging Face, Zenodo…).
PT, PTH, ONNX, H5, KERAS, TFLITE, ZIP · up to 10 MB, or a link

Track B · Cardiac Rhythm Recognition

Track B predictions
One row per requested ECG, probabilities between 0 and 1.
ecg_id,p_SR,p_AFIB,p_STACH,p_SBRAD,p_SARRH
CSV · up to 5 MB
Trained model weights & dependencies
Over 10 MB? Share a download link (Drive, Hugging Face, Zenodo…).
PT, PTH, ONNX, H5, KERAS, TFLITE, ZIP · up to 10 MB, or a link

Optional · Complete ECG Analysis

Demonstration (video or live link)optional
A short screen recording or a link to the running interface.
Link

Every team

Source code & inference instructions
Repository link (GitHub, GitLab, Hugging Face…) with installation and execution instructions.
Link
Technical description
Architecture and training procedure; preprocessing (sampling rate, filtering, normalisation); validation results and computational requirements.
PDF · up to 10 MB

Technical requirements

The official input consists of the ECG waveform only. Patient identifiers, demographic metadata and interpretation reports must not be used as model inputs for the main leaderboards.
Each CSV file contains one row per requested ECG and probability values between 0 and 1. The organiser-supplied validation script checks column order, duplicate identifiers and missing predictions.
The organisers provide a reference data loader, baseline notebook, scoring code and sample submission files. Finalist models must be executable offline.

Competition rules

1Participants must use PTB-XL version 1.0.3 and the official data partitions specified by the organisers.
2Patient-level separation between training, validation and test folds must be maintained.
3Test annotations must not be used for training, threshold selection or manual adjustment of predictions.
4All external datasets, pretrained models and relevant prior use of PTB-XL must be declared.
5Only ECG waveform inputs are permitted in the main leaderboard.
6Submissions must be reproducible using the source code and instructions provided.
7Results must be submitted before the official deadline.
8The organisers may inspect code and reproduce inference to verify submitted scores.

The use of publicly accessible labels for lookup or memorisation is prohibited.

The final evaluation manifest and rules will be communicated before the competition begins.

Timeline

Key dates.

Call for participation
November 2026
Registration
November – December 2026
Dataset and technical documentation release
13 November 2026
Model development
November – December 2026
Submission deadline
12 December 2026
Code verification and final evaluation
13 – 16 December 2026
Final presentations and demonstrations
17 December 2026
Results and awards
17 December 2026

The final dates, submission portal and evaluation environment will be confirmed by the organising committee.

Recognition

What’s at stake.

Best Overall Performance
Best Cardiac Abnormality Detection
Best Rhythm Recognition
Most Innovative Approach
Best Technical Presentation

Awards and certificates are subject to confirmation by the symposium committee.

Expected outcomes

Encourage research in cardiac signal analysis.
Introduce participants to multi-label medical classification.
Compare approaches to diagnostic and rhythm recognition.
Improve understanding of model errors and class imbalance.
Promote reproducible methods for ECG interpretation.
Identify promising techniques for future cardiology research.

Outstanding teams may be invited to present their methods and discuss research collaboration.

Why it matters

Automated ECG analysis can contribute to research on clinical decision-support systems, large-scale ECG review and accessible cardiac monitoring. The challenge also provides a common benchmark for studying signal quality, class imbalance and efficient inference.

By evaluating diagnostic statements and cardiac rhythms separately, the competition addresses two related but distinct tasks in ECG interpretation.

Important

This competition evaluates computational methods using annotated research data. Its results should not be interpreted as evidence of clinical readiness or diagnostic performance in routine care.

References

[1]Wagner, P., Strodthoff, N., Bousseljot, R.-D., Kreiseler, D., Lunze, F. I., Samek, W., & Schaeffter, T. (2020). PTB-XL, a Large Publicly Available Electrocardiography Dataset. Scientific Data, 7, 154.doi.org/10.1038/s41597-020-0495-6
[2]Wagner, P., Strodthoff, N., Bousseljot, R.-D., Samek, W., & Schaeffter, T. (2022). PTB-XL, a Large Publicly Available Electrocardiography Dataset (version 1.0.3). PhysioNet.doi.org/10.13026/kfzx-aw45
[3]Strodthoff, N., Wagner, P., Schaeffter, T., & Samek, W. (2021). Deep Learning for ECG Analysis: Benchmarks and Insights from PTB-XL. IEEE Journal of Biomedical and Health Informatics, 25(5), 1519–1528.
[4]PTB-XL dataset and annotation dictionary (PhysioNet).physionet.org/content/ptb-xl/1.0.3/
[5]Creative Commons Attribution 4.0 International Public License (CC BY 4.0).creativecommons.org/licenses/by/4.0/

ECGDetect-AI 2026 · ARC'26

Bring your best model.

ECGDetect-AI 2026 invites participants to develop methods for analysing clinical ECG signals and producing consistent predictions for diagnostic and rhythm labels.

The competition uses PTB-XL, a publicly released dataset of 21,799 ECG recordings from 18,869 patients.

Objectives

01Develop accurate methods for automatic analysis of 12-lead ECG recordings.
02Identify diagnostic abnormalities using multi-label classification.
03Recognise cardiac rhythms from ECG waveforms.
04Handle recordings with multiple findings and imbalanced label frequencies.
05Compare signal-processing methods and deep learning architectures.
06Explore models suitable for practical and efficient inference.
07Encourage reproducible research in AI-assisted electrocardiography.

Metric

Purpose

Macro AUROCPrimary

Measures discrimination across the five diagnostic superclasses; primary ranking metric.

Macro Average Precision (AP)

Assesses precision-recall performance, especially for less frequent labels.

Macro F1-score

Measures the balance between precision and recall after validation-based threshold selection.

Per-class sensitivity

Measures the proportion of positive recordings recovered for each diagnostic class.

Metric

Purpose

Macro F1-scorePrimary

Measures average positive-class F1 across the selected rhythm labels; primary ranking metric.

Macro AUROC

Evaluates the discrimination of rhythm probabilities.

Per-class recall

Measures the detection rate for each rhythm statement.

False-positive rate

Measures incorrect positive rhythm predictions.

Register your team, then upload each deliverable from the participant portal. Prediction files are checked the moment you upload them, and your latest upload before the deadline is the one evaluated.

Suggested approaches

Signal-processing features with classical classifiers
One-dimensional convolutional neural networks (1D CNNs)
Residual and Inception-style time-series networks
Transformer-based ECG architectures
Recurrent or state-space sequence models
Hybrid signal-processing and deep learning approaches

1Participants must use PTB-XL version 1.0.3 and the official data partitions specified by the organisers.
2Patient-level separation between training, validation and test folds must be maintained.
3Test annotations must not be used for training, threshold selection or manual adjustment of predictions.
4All external datasets, pretrained models and relevant prior use of PTB-XL must be declared.
5Only ECG waveform inputs are permitted in the main leaderboard.
6Submissions must be reproducible using the source code and instructions provided.
7Results must be submitted before the official deadline.
8The organisers may inspect code and reproduce inference to verify submitted scores.

The use of publicly accessible labels for lookup or memorisation is prohibited.

The final evaluation manifest and rules will be communicated before the competition begins.

Key dates.

Call for participation
November 2026
Registration
November – December 2026
Dataset and technical documentation release
13 November 2026
Model development
November – December 2026
Submission deadline
12 December 2026
Code verification and final evaluation
13 – 16 December 2026
Final presentations and demonstrations
17 December 2026
Results and awards
17 December 2026

The final dates, submission portal and evaluation environment will be confirmed by the organising committee.

What’s at stake.

Best Overall Performance
Best Cardiac Abnormality Detection
Best Rhythm Recognition
Most Innovative Approach
Best Technical Presentation

Awards and certificates are subject to confirmation by the symposium committee.

Expected outcomes

Encourage research in cardiac signal analysis.
Introduce participants to multi-label medical classification.
Compare approaches to diagnostic and rhythm recognition.
Improve understanding of model errors and class imbalance.
Promote reproducible methods for ECG interpretation.
Identify promising techniques for future cardiology research.

Outstanding teams may be invited to present their methods and discuss research collaboration.

12 leads. 21,799 ECGs. One model.

PTB-XL (version 1.0.3)

Choose your track.

Ship it from your team workspace.

Play fair, show your work.

Key dates.

What’s at stake.

Bring your best model.

ECGDetect-AI 2026

12 leads. 21,799 ECGs. One model.

PTB-XL (version 1.0.3)

Choose your track.

Ship it from your team workspace.

Play fair, show your work.

Key dates.

What’s at stake.

Bring your best model.