Leads selection during training and testing

[mohanad....@gmail.com]

Hello,

How is the selection of leads going to be during the training and testing?

Will the input be 12-lead ECG or selected 12-, 6-, 3-, and 2- ?

Do we have to run the training code 4 times for each leads scenario? or we have to run it only once and deal with the 12-leads then save models using these specific leads for testing using them only? 

regards,

Mohanad

[PhysioNet Challenge]

Dear Mohanad,

The training set only contains twelve-lead ECGs. The validation and test sets contain twelve-lead, six-lead, three-lead, and two-lead ECGs. Please see the data section of the Challenge webpage for the specific lead sets:
https://physionetchallenges.org/2021/#data

Your training code needs to train models for twelve-lead, six-lead, three-lead, and two-lead ECGs. Each lead set corresponds to a different Challenge, e.g., one team could win the two-lead Challenge while another team could win the three-lead Challenge, but each team’s trained models must be able to classify all four different lead sets.

The example code illustrates how we will run your code and how you might create your models:
https://github.com/physionetchallenges/matlab-classifier-2021
https://github.com/physionetchallenges/python-classifier-2021

You can run the train_model script to run your training code to create models for twelve-lead, six-lead, three-lead, and two-lead ECGs, and you can run the test_model script to run your trained models on these lead sets. Please note that the training code creates models for all four different lead sets, and the test code chooses the models dynamically from all four different lead sets.

You can use the extract_leads_wfdb script described in these repositories to create reduced-lead sets from the training data to make sure that your code is running as expected.

Best,
Matt

(On behalf of the Challenge team.)

https://PhysioNetChallenges.org/
https://PhysioNet.org/

Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email challenge at physionet.org. We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address. This email is maintained by a group. Please do not email us individually.

[had]

Can i use one model for all 4 tasks?

воскресенье, 28 февраля 2021 г. в 19:34:52 UTC+3, PhysioNet Challenge:

[physionet-challenges]

Dear Mohanad,

You are free to use identical or different models for the four tasks on the reduced validation and test sets. You should only make sure to keep the different name formats of the saved four models as indicated in the example code:

https://github.com/physionetchallenges/matlab-classifier-2021
https://github.com/physionetchallenges/python-classifier-2021

Best,

Nadi

(On behalf of the Challenge team)

---

https://physionetchallenges.github.io/2021/

www.PhysioNet.org

[Santiago Jiménez]

Dear Physionet organizers,

Maybe there is an error in the test_model.m file.

It allows to test models of 12, 6, 3 and 2 leads. But in the main for loop, it choose always to test the 12 leads models, since it reads the number of leads from the header file, and this is always compose of 12 leads.

Could you check it and correct me if I'm wrong?

Furthermore, I would like to ask you if it is possible to paralelize the testing script with a parfor loop. I made this change in the team_training_code.m file, where we all able to perform changes.

Thanks in advance for your response.

Best,

Santi

El dia dimecres, 3 de març de 2021 a les 0:23:43 UTC+1, physionet-challenges va escriure:

[physionet-challenges]

Dear Santi,

The test_model.m loads the data and detects the number of leads before the for loop and uses that information to test the data using the corresponding model for each lead set. If your test data contains only a 12-lead data set, only a 12-lead model will be used for the test. But if your test data set contains reduced lead data, other models will be used for testing the data.

We recommend the challengers to extract reduced-lead set from the Challenge 12-leads training data using the script, extract_leads_wfdb.py, that we provided in the following link:

https://github.com/physionetchallenges/python-classifier-2021#what-else-is-in-this-repository

You can then use the extracted reduced lead sets for testing which will result in the for loop calling and using other models (6-, 3- or 2-lead models).

You can make your changes to team_training_code.m and other scripts that you can edit and add your training and test code. We will look into using parfor loops for test and might update it if it helps the runtime. Please check the forum for more updates. 

Thanks for sharing.

Best,

Nadi

(On behalf of the Challenge team.)

https://physionetchallenges.github.io/

---

https://physionetchallenges.github.io/2021/

www.PhysioNet.org