MICCAI 2026 CARDIAC IMAGING THEME  ·  JOINTLY WITH THE STACOM WORKSHOP

// Guidelines

Official Challenge Rules

To ensure a fair, transparent, and reproducible challenge, all participants must strictly adhere to the following rules. Failure to comply may result in disqualification.

1
Registration & Teams: All individuals wishing to participate must register on Synapse using their real name, complete affiliation details, and institutional email address. Incomplete profiles may be removed without notice. Each team is permitted to have a maximum of six members.

Docker submission access: Test-set Docker submission access is granted to one designated contact per team, matching the Synapse Profile URL supplied on the Team Registration Form. This applies regardless of team size; only one member needs submission access, since the team's official entry is whichever container that person uploads. Teams wishing to change their designated submitter should email echorisk.miccai@gmail.com with the new Synapse Profile URL. Additional members beyond the designated submitter are not granted separate submission access.
2
Submission Requirements & Paper Eligibility: To be eligible for test-set evaluation, leaderboard placement, and awards, teams must submit both:
  • A methodology paper (4–8 pages, LNCS format) submitted via the EchoRisk OpenReview venue by 7 July 2026. Papers should describe the method and validation-set results.
  • A self-contained Docker image submitted via Synapse by 20 August 2026. Test-set access will only be granted to teams with a valid paper submission.

This policy ensures scientific accountability and is consistent with major MICCAI challenges such as HECKTOR and BraTS. Teams that submit a Docker container without a valid paper will not appear on the official leaderboard and are not eligible for awards.

3
Publication Agreement: By participating, you agree that your submitted methodology paper may be made publicly available on the challenge website. Organizers reserve the right to use information provided by participants, including evaluation scores, predicted labels, and paper content, for the joint challenge summary publication.
4
Account Integrity: Participants are strictly prohibited from registering multiple teams or accounts. Creating alternative accounts to bypass submission limits will lead to immediate disqualification of all associated accounts.
5
Data Usage (DUA): Redistribution, sharing, or transfer of the CARDIOCARE challenge data or data links is strictly prohibited. Participants must use the data solely for their own model development within the context of this specific challenge.
6
Methods & External Data: Solutions must be fully automated during inference with no manual interventions (e.g., manual annotation of test cases). The use of public external datasets and pre-trained models (e.g., EchoNet-Dynamic, foundation models) is permitted, provided they are publicly available to all and explicitly disclosed in the methodology paper.

Cross-task data use (Task 3): Ground-truth Task 1 or Task 2 labels must not be used as direct inputs to a Task 3 pipeline, at any timepoint, for any patient. Auxiliary models may be trained on Task 1 or Task 2 public training data, or any other publicly available dataset, to predict LVEF or LV dysfunction from DICOM video, and those predictions may be used as inputs to Task 3. Auxiliary model input must be restricted to the DICOM video itself; standard temporal modelling within a single cine loop is unrestricted, but the model must not have access to timepoint, visit order, or patient trajectory information across visits. Auxiliary model architecture, training data source, and exact input schema must be disclosed in the methodology paper. Submissions with disclosed methodology inconsistent with observed inference behaviour may be excluded from the joint summary paper at the organisers' discretion.

// Q&A

Frequently Asked Questions

Do I have to participate in all tasks?
No. You may choose to participate in Task 1, Task 2, Task 3, or any combination of them. Each task has its own independent leaderboard, evaluation, and prize pool.
Can I use external data or pretrained models?
Yes. You may use publicly available external datasets and pretrained weights (e.g. EchoNet, EchoPrime, ImageNet). Fine-tuning a public checkpoint on the challenge training data is also permitted. However, fine-tuning on a private or institutional dataset is not permitted — all external resources must be publicly available so that any team could in principle reproduce your pipeline. All external data and weights used must be explicitly disclosed in your methodology paper, including any additional fine-tuning performed and on what data.
Is a paper submission required to access the test set?
Yes. A valid methodology paper submitted via the EchoRisk OpenReview venue by 7 July 2026 is a prerequisite for test-set evaluation and leaderboard inclusion. This is a firm requirement, consistent with major MICCAI challenges such as HECKTOR and BraTS.

The paper submitted by 7 July should describe your approach and validation-set results as they stand at that point. The camera-ready version, due 13 August 2026, may incorporate minor refinements and additional validation-set results. The core method must remain substantially the same as the reviewed version. Whether final test-set results can be included in the camera-ready will depend on evaluation timeline and will be communicated to all teams after the Docker deadline closes.
What are the computational resources for evaluation?
Submissions will be evaluated using one dedicated GPU (≥16 GB VRAM), multi-core CPU resources, and ≥32 GB of system memory. Final hardware specifications (exact GPU model, CPU configuration, RAM) will be announced no later than 3 months before the test submission deadline.

Each task is allocated a maximum wall-clock inference time of 15 minutes per task. Submissions that exceed this limit will be marked as failed for the affected task. Containers are evaluated on the largest case first; failure on the first case will terminate evaluation for that submission.

Note: All submissions must complete inference within the allocated resources. External API calls and internet access are strictly prohibited during inference.
Is there an online leaderboard during development?
No. Training and validation sets are released with ground-truth labels, allowing participants to conduct their own evaluation locally. Official rankings are determined solely by organiser evaluation of the hidden test set submissions.
Do we need to upload our training code?
No, we do not require teams to upload their training code. Only your inference code packaged as a self-contained Docker image is needed for challenge participation and final evaluation.
Are organizers allowed to participate?
Members of the organizer institutes are welcome to submit models and participate in the challenge for scientific benchmarking, but they are not eligible for monetary awards and will not be ranked on the public leaderboard.
Who should I contact with questions?
For any questions regarding registration, data access, or submission, contact the organising team at echorisk.miccai@gmail.com. We aim to respond within 3 working days.
What is the maximum number of team members allowed?
Each participating team can consist of up to a maximum of 6 people.
Can we preprocess the data?
Yes. During training, participants are allowed to perform standard data preprocessing. We do not restrict any data preprocessing techniques—we only evaluate your final model predictions.
Are frame rate and pixel spacing metadata available in the DICOM files?
No. Temporal calibration tags (FrameTime, CineRate) and spatial calibration tags (SequenceOfUltrasoundRegions, PixelSpacing) are absent from all released DICOM files. This is a known characteristic of the dataset resulting from the anonymisation pipeline applied prior to data transfer and affects all tasks, splits, and sites consistently.

NumberOfFrames is present in every file and can be read directly from the DICOM header. Participants should develop methods that do not assume physical temporal or spatial calibration is available at inference. This is consistent with established echo AI benchmarks including EchoNet-Dynamic, and reflects realistic deployment conditions in heterogeneous clinical environments.
Can the A2C or A4C view be missing at test time?
Yes. The dataset documentation notes that either the A2C or A4C DICOM may be absent for some visits. This reflects real clinical acquisition variability and applies across all splits.

Participants should treat the validation set as representative of test-set conditions with respect to view availability. Your pipeline must handle missing views gracefully. We recommend characterising missingness in the validation set locally to inform your design decisions. Per-split missingness statistics will not be released prior to test-set evaluation.
Does our paper need to be accepted to appear on the leaderboard?
No. Leaderboard eligibility is determined by a desk check, not peer review outcome. Any team that submits a valid paper by 7 July 2026 and passes the desk check will be eligible for test-set evaluation and leaderboard inclusion, regardless of whether the paper is accepted into the workshop proceedings. Peer review runs in parallel and determines proceedings inclusion only.

The desk check criteria are:
  1. Minimum 4 pages in LNCS format (excluding references)
  2. Contains a clearly identifiable Methods section describing the algorithm
  3. Reports at least one quantitative result on the validation set for the task(s) entered
  4. At least one author is a registered team member on Synapse
Desk check results will be communicated within two weeks of the paper deadline.

There is no cap on the number of paper co-authors. The six-member cap applies to the registered Synapse team only. Co-authors not registered as team members are not eligible for awards.
Who qualifies for the joint challenge summary paper?
The top 3 ranked teams per task will be invited to contribute to the joint challenge summary paper. Eligibility is based solely on final leaderboard rank and compliance with challenge rules and submission requirements. No subjective criteria such as technical novelty or code quality are applied.

Each qualifying team may nominate up to three co-authors. Given that Tasks 1, 2, and 3 have independent leaderboards, a team performing well across multiple tasks may be invited to contribute to more than one task's summary.
Does the methodology paper need to describe the exact final model?
No. The paper submitted by 7 July describes your approach as it stands at that point. Your model may continue to evolve after the paper deadline. The Docker container submitted by 20 August 2026 represents your final method and is what gets evaluated on the test set.

The camera-ready version, due 13 August 2026, may incorporate minor refinements and additional validation-set results. However, the core method must remain substantially the same as the reviewed version. Wholesale changes to the approach between the initial submission and the camera-ready are not permitted, as peer review is conducted on the initial version.
Are there required sections in the methodology paper?
No fixed structure is mandated beyond the LNCS format and 4 to 8 page requirement. We recommend the following structure as it covers the minimum content reviewers will expect:
  • Introduction: brief motivation and problem statement
  • Methods: description of your algorithm, architecture, and any external data or pretrained models used
  • Experiments: validation-set results using the official challenge metrics; this is a mandatory content requirement for desk check eligibility
  • Discussion / Conclusion: limitations and key findings
Disclosure of any external datasets or pretrained models used is mandatory and must appear in the paper per challenge Rule 6.
Can more than one team member get submission access?
No. Docker submission access is granted to one designated contact per team, matching the Synapse Profile URL on your Team Registration Form. Only one member needs access, since your team's official entry is whichever container that person uploads on the team's behalf. To change who this is, email echorisk.miccai@gmail.com with the new Synapse Profile URL.
Where do I find step-by-step Docker submission instructions?
See the dedicated Submission Guide page, which covers Synapse certification requirements, locating your evaluation queue, the exact Docker push and submit sequence, and troubleshooting for the most common errors teams encounter.

Ready to participate?

Register on Synapse →

Questions? echorisk.miccai@gmail.com

CARDIOCARE
FORTH
UOI
HMU
NKUA
BOCOC
IEO
KSBC