Help with reproduction of results

[HReynaud]

Hello,

Thank you for the extremely valuable data you released.

I am currently working with it and wanted to reproduce the results from your paper, as I have been unable to achieve good scores (R2/MAE) by re-implementing.

A few code things that could be fixed:

• In setup.py the package sklearn should be scikit-learn, otherwise pip crashes and the echonet package does not get installed.
• echo.py does not load the data properly because the splits in the FileList.csv files use fold indices instead of "TRAIN", "VAL", "TEST".
• It looks like echo.py can only handle one view at a time right now.

Higher level comments:

• In the paper you mention that you can train on paired PSAX and A4C videos for patients who got both scans. How can this be reproduced with the public dataset ? Is there any way to match videos from the PSAX view and A4C view ?
• It would be great if the code allowed to reproduce the paper experiments. I can see many scripts that seem to be partially doing this, but it lacks a good readme file for easy reproduction, while following the format of the released dataset.
• The released dataset zip file has an unnecessary long list of nested folders Echonet-Peds/echonetpediatric/pediatric_echo_avi/pediatric_echo_avi/{view}/Videos/{video_name}.avi. If possible, making the folder structure similar to Echonet-Dynamic would be welcome: Echonet-Peds/{view}/Videos/{video_name}.avi

Again, thank you for your hard work, I hope these comments can help make this work easier to extend upon.