Confusions on Signal Trimming and ECG Layout Formats for PhysioNet Challenge Evaluation

[申润泽]

Dear Challenge Organizers,

Hello!

During the evaluation process of the PhysioNet Challenge, I encountered some confusion and would appreciate your assistance.

Question 1: According to the official emails and documentation, I understand that during the evaluation, you only compare the segments of the ECG time series that were plotted on the ECG image. For example, you mentioned:

• Leads I, II, III: 0 to 2.5 seconds
• Leads aVR, aVL, aVF: 2.5 to 5 seconds
• Leads V1, V2, V3: 5 to 7.5 seconds
• Leads V4, V5, V6: 7.5 to 10 seconds

However, when I follow the official code, each lead's signal remains a complete 10-second signal. I do not see any operation that trims the 10-second signal to 2.5 seconds for specific leads during the evaluation.

Could you please clarify whether the official evaluation code includes steps to trim each lead's 10-second signal to 2.5 seconds? If so, could you help me understand how this is implemented or point out where this operation is performed? Additionally, does the official code modify the .dat files in any specific way? Because when I plot the signals, each lead still shows 10 seconds.

Question 2: I would like to know if the hidden dataset's ECGs follow a uniform 3x4 layout format, or if there are different formats. For instance, could there be ECGs arranged in a 13x1 format, or is it always a 3x4 format but with 10-second long leads?

Thank you very much for your assistance!

Sincerely,

Runze, Shen

[PhysioNet Challenge]

Dear Runze, Shen,

For training, we are providing the full ECG time series (the .hea and .dat files). You may or may not want to use the segments of the ECG time series that are not plotted in the provided ECG images (the .png files). For example, you may want to generate new ECG image files with different layouts to train your model on different ECG image layouts.

For inference, we are only asking you to digitize the ECG segments that were plotted in the ECG images. If you want to produce new ECG time series that mask the unplotted ECG segments, then you can run the following command from the ECG image generation code; the --mask_unplotted_samples argument masks the unplotted samples:

 python gen_ecg_images_from_data_batch.py \
     -i ptb-xl/records500/00000 \
     -o ptb-xl/records500_hidden/00000 \
     --print_header \
     --mask_unplotted_samples

This command will produce data that look like the validation and test set. You can run the commands on this section of the webpage to generate data that are formatted in very much the same way as the validation and test sets:
https://physionetchallenges.org/2024/#accessing-data

The 3x4 layout with 12 short ECG segments and 1 long ECG segment is a standard format for ECG images, and I would expect many of the ECG images in the validation and test sets to have this format, but there are other formats, and the hidden data may include some of them. For example, the Challenge webpage contains at least one example of a different format.

Best,
Matt
(On behalf of the Challenge team.)

Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at physionetchallenge.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.