![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F23631%2F72b7207c46abbd1975e47f1e7b415723%2FScreenshot_20251218_102115.png?generation=1766049716902054&alt=media\")
For example, in sample 101402388 we see the following signal at the end of the V6 lead: […, 0.018, 0.004, -0.045], aka \"sudden drop\", visualized here:\n
This is not even physiological and makes no sense.
\nBut the real training image does not contain this drop:\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F23631%2Fce4dee9d207ca34d273336ae1234957f%2FScreenshot_20251218_102222.png?generation=1766049750436887&alt=media\")
I suspect that a similar problem is also present in the test data.
", "messages": [ { "id": "3378469", "postDate": "12/18/2025 09:23:22", "content": "For example, in sample 101402388 we see the following signal at the end of the V6 lead: […, 0.018, 0.004, -0.045], aka \"sudden drop\", visualized here:\n
This is not even physiological and makes no sense.
\nBut the real training image does not contain this drop:\n
I suspect that a similar problem is also present in the test data.
", "rawMarkdown": "For example, in sample 101402388 we see the following signal at the end of the V6 lead: [..., 0.018, 0.004, -0.045], aka \"sudden drop\", visualized here:\n\n\nThis is not even physiological and makes no sense.\n\nBut the real training image does not contain this drop:\n\n\nI suspect that a similar problem is also present in the test data.", "votes": null }, { "id": "3379484", "postDate": "12/20/2025 00:12:36", "content": "There seem to be several such cases actually, like 1103158012 and 1259907878 for example, there are numerous others to, I think its best to just ignore them
", "rawMarkdown": "There seem to be several such cases actually, like 1103158012 and 1259907878 for example, there are numerous others to, I think its best to just ignore them", "votes": null }, { "id": "3380239", "postDate": "12/22/2025 02:36:31", "content": "The \"drops\" in the last sample of the signal are artifacts caused by resampling of the data. Different ECG machines acquire ECG signals at different sampling rates, and different resampling and filtering algorithms on data from these devices (as well as on the devices themselves) can create artifacts in the signal. Moreover, the \"effective\" sampling rates of the ECG images resulting from the plotting program, the printer DPI, and the scanner or camera PPI is lower than the sampling rates of the signals, so the process of printing and digitizing the ECG images is inherently lossy.
\nA small drop in a single sample or time point of some records has a negligible impact on scores, affects all teams equally, and, importantly, is an example of the many artifacts that complicate ECG digitization. Please see these posts for examples.
\nThe field of signal processing studies these and many related issues for ECG signals and other biomedical and non-biomedical signals.
", "rawMarkdown": "The \"drops\" in the last sample of the signal are artifacts caused by resampling of the data. Different ECG machines acquire ECG signals at different sampling rates, and different resampling and filtering algorithms on data from these devices (as well as on the devices themselves) can create artifacts in the signal. Moreover, the \"effective\" sampling rates of the ECG images resulting from the plotting program, the printer DPI, and the scanner or camera PPI is lower than the sampling rates of the signals, so the process of printing and digitizing the ECG images is inherently lossy. \n\nA small drop in a single sample or time point of some records has a negligible impact on scores, affects all teams equally, and, importantly, is an example of the many artifacts that complicate ECG digitization. Please see these [posts](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/613179) [for](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/614972) [examples](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/614416).\n\nThe field of signal processing studies these and many related issues for ECG signals and other biomedical and non-biomedical signals.", "votes": null } ], "comments": [ { "id": 3379484, "author_name": "henrychibueze", "author_url": "", "post_date": "12/20/2025 00:12:36", "content": "There seem to be several such cases actually, like 1103158012 and 1259907878 for example, there are numerous others to, I think its best to just ignore them
", "votes": null, "replies": [ { "id": 3380239, "author_name": "matthewreyna", "author_url": "", "post_date": "12/22/2025 02:36:31", "content": "The \"drops\" in the last sample of the signal are artifacts caused by resampling of the data. Different ECG machines acquire ECG signals at different sampling rates, and different resampling and filtering algorithms on data from these devices (as well as on the devices themselves) can create artifacts in the signal. Moreover, the \"effective\" sampling rates of the ECG images resulting from the plotting program, the printer DPI, and the scanner or camera PPI is lower than the sampling rates of the signals, so the process of printing and digitizing the ECG images is inherently lossy.
\nA small drop in a single sample or time point of some records has a negligible impact on scores, affects all teams equally, and, importantly, is an example of the many artifacts that complicate ECG digitization. Please see these posts for examples.
\nThe field of signal processing studies these and many related issues for ECG signals and other biomedical and non-biomedical signals.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3378469": "For example, in sample 101402388 we see the following signal at the end of the V6 lead: [..., 0.018, 0.004, -0.045], aka \"sudden drop\", visualized here:\n\n\nThis is not even physiological and makes no sense.\n\nBut the real training image does not contain this drop:\n\n\nI suspect that a similar problem is also present in the test data.", "3379484": "There seem to be several such cases actually, like 1103158012 and 1259907878 for example, there are numerous others to, I think its best to just ignore them", "3380239": "The \"drops\" in the last sample of the signal are artifacts caused by resampling of the data. Different ECG machines acquire ECG signals at different sampling rates, and different resampling and filtering algorithms on data from these devices (as well as on the devices themselves) can create artifacts in the signal. Moreover, the \"effective\" sampling rates of the ECG images resulting from the plotting program, the printer DPI, and the scanner or camera PPI is lower than the sampling rates of the signals, so the process of printing and digitizing the ECG images is inherently lossy. \n\nA small drop in a single sample or time point of some records has a negligible impact on scores, affects all teams equally, and, importantly, is an example of the many artifacts that complicate ECG digitization. Please see these [posts](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/613179) [for](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/614972) [examples](https://www.kaggle.com/competitions/physionet-ecg-image-digitization/discussion/614416).\n\nThe field of signal processing studies these and many related issues for ECG signals and other biomedical and non-biomedical signals." }, "source": "meta" }