Welcome to the Extreme Cardiac MRI Analysis Challenge under Respiratory Motion (CMRxMotion). This a medical image analysis challenge officially registered in MICCAI 2022.

Registration of CMRxMotion Challenge is open from 15 April 2022.


The CMR imaging quality is susceptible to respiratory motion artefacts. The aim of this challenge is to assess the effects of respiratory motion on CMR imaging quality and exam the robustness of automatic segmentation models under different levels of respiratory motion.


Cardiac magnetic resonance (CMR) imaging is the current gold-standard modality for the evaluation of cardiac structure and function. Machine learning-based approaches have achieved remarkable performance in previous CMR challenges (e.g., ACDC, M&Ms). However, in clinical practice, the model performance is challenged by inconsistent imaging environments (e.g., vendors and protocols), population shifts (normal v.s. pathological cases) and unexpected human behaviours (e.g., body movements). It is useful to investigate potential failure modes by exposing a trained machine learning model to extreme cases in a ‘stress test’.

To date, existing challenges on model generalisability focus on the vendor variability and anatomical structure variations while the implications of human behaviours are less explored. For CMR acquisition, respiration motion is one of the major problems. Patients with acute symptoms cannot adhere to the breath-hold instructions, resulting in degraded image quality and inaccurate analysis.

CMR acquisition design

In this challenge, we introduce the CMRxMotion dataset, a realistic cardiac MRI dataset including extreme cases with different levels of respiratory motions. We aim to provide such an extreme CMR dataset that allows for a more comprehensive evaluation of model robustness under respiratory motion.

We attacked the standard of procedure (SOP) by redesigning the participant’s behaviours during acquisition. For 45 healthy volunteers, we perform clinical CMR scans using the same MRI scanner (Siemens 3T MRI scanner MAGNETOM Vida). Volunteers are trained to act in 4 manners, respectively: a) adhere to the breath-hold instructions; b) halve the breath-hold period; c) breathe freely; and d) breathe intensively.

Therefore, for a single volunteer, we obtain a set of paired CMR images under 4 levels of respiratory motion. Radiologists first assess the image quality and recognise the images with bad quality. For those images with diagnostic quality, radiologists further segment the left ventricle, left ventricle myocardium and right ventricle. We release the images, quality scores and ground-truth segmentations (if any) of 25 volunteers for training and validation. The remaining 20 volunteers will be withheld for the test.

Challenge tasks

Based on this extreme cardiac dataset, we design two tasks: 1) the image quality assessment task and 2) the segmentation task.

  • Task 1: challenge participants are expected to develop a image quality assessment model for CMR.
  • Task 2: participants are expected to develop segmentation models that are robust to respiratory motion artifacts.

Top 3 winners of each task receive monetary awards:1st place $500, 2nd place $200, 3rd place $100.


The schedule of the challenge is as follows. All deadlines are London time (UTC/GMT +0:00).

15-Aprwebsite opens for registration
release training cases (image, quality label and segmentation)
01-Julrelease validation cases (image)
06-Julvalidation data evaluating system opens (Leaderboard and submission system are available.)
15-Juldeadline for STACOM placeholder paper submission
15-Augregistration deadline, submission system opens for test set
05-Sepdocker submission deadline
18-Seprelease final results at STACOM workshop


To contact us by email, send to CMRxMotion@163.com or CMRxMotion@gmail.com.


We are looking for sponsors for this challenge.

If you are interested, please contact: shuowang@fudan.edu.cn