{
  "id": 444265,
  "title": "Active Extravasation on Arterial / PV phase",
  "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/444265",
  "author_name": "Reuben Schmidt",
  "post_date": "2023-10-01T06:40:14",
  "votes": 15,
  "comment_count": 1,
  "views": 0,
  "content": "<p>In the train data, 1564 patients (49.6%) have 2 series IDs, indicating arterial and portal venous phase studies. The remaining 1583 patients (50.4%) have 1 series ID, which could be either phase. As noted in <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427226\" target=\"_blank\">this thread</a>, if a patient has 2 series IDs, the series ID with the higher aortic HU value (in train_series_meta.csv) is arterial phase.</p>\n<p>Briefly, for abdo trauma CT, IV contrast is always given. 20 seconds after it is given a 3D CT is taken in \"arterial phase\", when contrast has reached arteries. After about another minute, when the contrast reaches veins, another CT stack is taken in \"portal venous phase\".</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F8462229b8f968d4fb37ccfab2dae61d7%2F10127_53380_0448.png?generation=1695339037288269&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F45879314a587ca972bbb794eb0a957fb%2F10127_1554_0448.png?generation=1695339025077302&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>10127/53380/0448.dcm</td>\n<td>10127/1554/0448.dcm</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nThe most important label in the competition is extravasation, weighted 6x. Extravasation (or extrav as it gets called in the hospital) means contrast leaking from a ruptured blood vessel into surrounding tissue. Arterial bleeds show extrav in arterial phase (high aortic HU). Venous bleeds show extrav in portal venous phase (low aortic HU). Sometimes we will see extrav on both. </p>\n<p>Here is an example of a venous bleed, with extravasation only in portal venous phase:</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F919e3cb55e957f610217485ed3dbf216%2Fdownload.png?generation=1695346962213651&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F06254287429b6dd666e5debb860470d6%2Fdownload%20(1).png?generation=1695346950386203&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>11925/33599/490.dcm</td>\n<td>11925/60003/490.dcm</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nHere is an example of an arterial bleed, with extrav in both arterial and portal venous phase:</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fcb00baed39ebb6d3e64d7b741f0e9300%2Fdownload%20(2).png?generation=1695405186821176&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fa76bb5895c84a1b1d52b62c54d9b59c7%2Fdownload%20(1).png?generation=1695405225655261&amp;alt=media\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nIn real life, we more often than not see extrav on both phases. I decided to check for associations between number of series and extrav in this dataset. </p>\n<table>\n<thead>\n<tr>\n<th></th>\n<th>Pts who have one series_id (1583)</th>\n<th>Pts who have two series_ids (1564)</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Pts who have active extravasation</td>\n<td>100 (6.3%)</td>\n<td>100 (6.4%)</td>\n</tr>\n<tr>\n<td>Proportion who have active extravasation on one series</td>\n<td>100%</td>\n<td>41 (41%)</td>\n</tr>\n<tr>\n<td>Proportion who have active extravasation on both series</td>\n<td>N/A</td>\n<td>59 (59%)</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nIf a patient has two series_ids, but active extravasation only on one, which is more likely - portal venous or arterial phase?

</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>20 (49%)</td>\n<td>21 (51%)</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nAttached is the code I used to work this out. Thanks to Harshit Sheoran for fact-checking earlier versions of this.</p>\n<p>In summary, patients with two series are more likely to have extravasation on both series. If extravasation is only on one of two series, the dataset appears to evenly represent both arterial and venous bleeds.</p>\n<p>Hope that helps with some people's exploration. </p>",
  "messages": [
    {
      "id": 2463253,
      "postDate": "2023-10-01T06:40:14Z",
      "content": "<p>In the train data, 1564 patients (49.6%) have 2 series IDs, indicating arterial and portal venous phase studies. The remaining 1583 patients (50.4%) have 1 series ID, which could be either phase. As noted in <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427226\" target=\"_blank\">this thread</a>, if a patient has 2 series IDs, the series ID with the higher aortic HU value (in train_series_meta.csv) is arterial phase.</p>\n<p>Briefly, for abdo trauma CT, IV contrast is always given. 20 seconds after it is given a 3D CT is taken in \"arterial phase\", when contrast has reached arteries. After about another minute, when the contrast reaches veins, another CT stack is taken in \"portal venous phase\".</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F8462229b8f968d4fb37ccfab2dae61d7%2F10127_53380_0448.png?generation=1695339037288269&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F45879314a587ca972bbb794eb0a957fb%2F10127_1554_0448.png?generation=1695339025077302&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>10127/53380/0448.dcm</td>\n<td>10127/1554/0448.dcm</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nThe most important label in the competition is extravasation, weighted 6x. Extravasation (or extrav as it gets called in the hospital) means contrast leaking from a ruptured blood vessel into surrounding tissue. Arterial bleeds show extrav in arterial phase (high aortic HU). Venous bleeds show extrav in portal venous phase (low aortic HU). Sometimes we will see extrav on both. </p>\n<p>Here is an example of a venous bleed, with extravasation only in portal venous phase:</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F919e3cb55e957f610217485ed3dbf216%2Fdownload.png?generation=1695346962213651&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F06254287429b6dd666e5debb860470d6%2Fdownload%20(1).png?generation=1695346950386203&amp;alt=media\" alt=\"\"></td>\n</tr>\n<tr>\n<td>11925/33599/490.dcm</td>\n<td>11925/60003/490.dcm</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nHere is an example of an arterial bleed, with extrav in both arterial and portal venous phase:</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fcb00baed39ebb6d3e64d7b741f0e9300%2Fdownload%20(2).png?generation=1695405186821176&amp;alt=media\" alt=\"\"></td>\n<td><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fa76bb5895c84a1b1d52b62c54d9b59c7%2Fdownload%20(1).png?generation=1695405225655261&amp;alt=media\" alt=\"\"></td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nIn real life, we more often than not see extrav on both phases. I decided to check for associations between number of series and extrav in this dataset. </p>\n<table>\n<thead>\n<tr>\n<th></th>\n<th>Pts who have one series_id (1583)</th>\n<th>Pts who have two series_ids (1564)</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Pts who have active extravasation</td>\n<td>100 (6.3%)</td>\n<td>100 (6.4%)</td>\n</tr>\n<tr>\n<td>Proportion who have active extravasation on one series</td>\n<td>100%</td>\n<td>41 (41%)</td>\n</tr>\n<tr>\n<td>Proportion who have active extravasation on both series</td>\n<td>N/A</td>\n<td>59 (59%)</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nIf a patient has two series_ids, but active extravasation only on one, which is more likely - portal venous or arterial phase?

</p>\n<table>\n<thead>\n<tr>\n<th>Arterial phase</th>\n<th>Portal venous phase</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>20 (49%)</td>\n<td>21 (51%)</td>\n</tr>\n</tbody>\n</table>\n<p>.<br>\nAttached is the code I used to work this out. Thanks to Harshit Sheoran for fact-checking earlier versions of this.</p>\n<p>In summary, patients with two series are more likely to have extravasation on both series. If extravasation is only on one of two series, the dataset appears to evenly represent both arterial and venous bleeds.</p>\n<p>Hope that helps with some people's exploration. </p>",
      "rawMarkdown": "In the train data, 1564 patients (49.6%) have 2 series IDs, indicating arterial and portal venous phase studies. The remaining 1583 patients (50.4%) have 1 series ID, which could be either phase. As noted in [this thread](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427226), if a patient has 2 series IDs, the series ID with the higher aortic HU value (in train_series_meta.csv) is arterial phase.\n\nBriefly, for abdo trauma CT, IV contrast is always given. 20 seconds after it is given a 3D CT is taken in \"arterial phase\", when contrast has reached arteries. After about another minute, when the contrast reaches veins, another CT stack is taken in \"portal venous phase\".\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F8462229b8f968d4fb37ccfab2dae61d7%2F10127_53380_0448.png?generation=1695339037288269&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F45879314a587ca972bbb794eb0a957fb%2F10127_1554_0448.png?generation=1695339025077302&alt=media) |\n| 10127/53380/0448.dcm | 10127/1554/0448.dcm |\n   \n.\nThe most important label in the competition is extravasation, weighted 6x. Extravasation (or extrav as it gets called in the hospital) means contrast leaking from a ruptured blood vessel into surrounding tissue. Arterial bleeds show extrav in arterial phase (high aortic HU). Venous bleeds show extrav in portal venous phase (low aortic HU). Sometimes we will see extrav on both. \n\nHere is an example of a venous bleed, with extravasation only in portal venous phase:\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F919e3cb55e957f610217485ed3dbf216%2Fdownload.png?generation=1695346962213651&alt=media)  | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F06254287429b6dd666e5debb860470d6%2Fdownload%20(1).png?generation=1695346950386203&alt=media)  |\n| 11925/33599/490.dcm | 11925/60003/490.dcm |\n\n\n.\nHere is an example of an arterial bleed, with extrav in both arterial and portal venous phase:\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fcb00baed39ebb6d3e64d7b741f0e9300%2Fdownload%20(2).png?generation=1695405186821176&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fa76bb5895c84a1b1d52b62c54d9b59c7%2Fdownload%20(1).png?generation=1695405225655261&alt=media) |\n\n\n.\nIn real life, we more often than not see extrav on both phases. I decided to check for associations between number of series and extrav in this dataset. \n\n\n| | Pts who have one series_id (1583)  | Pts who have two series_ids (1564)  |\n| --- | --- | --- |\n| Pts who have active extravasation | 100 (6.3%) | 100 (6.4%) |\n| Proportion who have active extravasation on one series | 100% | 41 (41%) |\n| Proportion who have active extravasation on both series | N/A | 59 (59%) |\n\n\n.\nIf a patient has two series_ids, but active extravasation only on one, which is more likely - portal venous or arterial phase?

\n\n| Arterial phase | Portal venous phase |\n| --- | --- |\n| 20 (49%) | 21 (51%) |\n\n\n.\nAttached is the code I used to work this out. Thanks to Harshit Sheoran for fact-checking earlier versions of this.\n\nIn summary, patients with two series are more likely to have extravasation on both series. If extravasation is only on one of two series, the dataset appears to evenly represent both arterial and venous bleeds.\n\nHope that helps with some people's exploration. ",
      "votes": 15
    },
    {
      "id": 2467207,
      "postDate": "2023-10-04T10:56:46.297Z",
      "content": "<p>Did you plan to use it in solution notebook?</p>",
      "rawMarkdown": "Did you plan to use it in solution notebook?"
    }
  ],
  "comments": [
    {
      "id": 2467207,
      "author_name": "doxon johnson",
      "author_url": "",
      "post_date": "2023-10-04T10:56:46.297000",
      "content": "<p>Did you plan to use it in solution notebook?</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2463253": "In the train data, 1564 patients (49.6%) have 2 series IDs, indicating arterial and portal venous phase studies. The remaining 1583 patients (50.4%) have 1 series ID, which could be either phase. As noted in [this thread](https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/427226), if a patient has 2 series IDs, the series ID with the higher aortic HU value (in train_series_meta.csv) is arterial phase.\n\nBriefly, for abdo trauma CT, IV contrast is always given. 20 seconds after it is given a 3D CT is taken in \"arterial phase\", when contrast has reached arteries. After about another minute, when the contrast reaches veins, another CT stack is taken in \"portal venous phase\".\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F8462229b8f968d4fb37ccfab2dae61d7%2F10127_53380_0448.png?generation=1695339037288269&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F45879314a587ca972bbb794eb0a957fb%2F10127_1554_0448.png?generation=1695339025077302&alt=media) |\n| 10127/53380/0448.dcm | 10127/1554/0448.dcm |\n   \n.\nThe most important label in the competition is extravasation, weighted 6x. Extravasation (or extrav as it gets called in the hospital) means contrast leaking from a ruptured blood vessel into surrounding tissue. Arterial bleeds show extrav in arterial phase (high aortic HU). Venous bleeds show extrav in portal venous phase (low aortic HU). Sometimes we will see extrav on both. \n\nHere is an example of a venous bleed, with extravasation only in portal venous phase:\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F919e3cb55e957f610217485ed3dbf216%2Fdownload.png?generation=1695346962213651&alt=media)  | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2F06254287429b6dd666e5debb860470d6%2Fdownload%20(1).png?generation=1695346950386203&alt=media)  |\n| 11925/33599/490.dcm | 11925/60003/490.dcm |\n\n\n.\nHere is an example of an arterial bleed, with extrav in both arterial and portal venous phase:\n\n| Arterial phase  | Portal venous phase |\n| --- | --- |\n| ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fcb00baed39ebb6d3e64d7b741f0e9300%2Fdownload%20(2).png?generation=1695405186821176&alt=media) | ![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4271956%2Fa76bb5895c84a1b1d52b62c54d9b59c7%2Fdownload%20(1).png?generation=1695405225655261&alt=media) |\n\n\n.\nIn real life, we more often than not see extrav on both phases. I decided to check for associations between number of series and extrav in this dataset. \n\n\n| | Pts who have one series_id (1583)  | Pts who have two series_ids (1564)  |\n| --- | --- | --- |\n| Pts who have active extravasation | 100 (6.3%) | 100 (6.4%) |\n| Proportion who have active extravasation on one series | 100% | 41 (41%) |\n| Proportion who have active extravasation on both series | N/A | 59 (59%) |\n\n\n.\nIf a patient has two series_ids, but active extravasation only on one, which is more likely - portal venous or arterial phase?

\n\n| Arterial phase | Portal venous phase |\n| --- | --- |\n| 20 (49%) | 21 (51%) |\n\n\n.\nAttached is the code I used to work this out. Thanks to Harshit Sheoran for fact-checking earlier versions of this.\n\nIn summary, patients with two series are more likely to have extravasation on both series. If extravasation is only on one of two series, the dataset appears to evenly represent both arterial and venous bleeds.\n\nHope that helps with some people's exploration. ",
    "2467207": "Did you plan to use it in solution notebook?"
  }
}