{
  "id": 344862,
  "title": "Train set scan to ignore",
  "url": "/competitions/rsna-2022-cervical-spine-fracture-detection/discussion/344862",
  "author_name": "Sohier Dane",
  "post_date": "2022-08-16T21:30:17.721000",
  "votes": 38,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The scan <code>1.2.826.0.1.3680043.20574</code> does not include a full cervical spine and should be ignored. Thanks to <a href=\"https://www.kaggle.com/kretes\" target=\"_blank\">@kretes</a> <a href=\"https://www.kaggle.com/code/kretes/pixdim-spacing-distributions?scriptVersionId=102885798#1.2.826.0.1.3680043.20574\" target=\"_blank\">for catching this</a>.</p>",
  "messages": [
    {
      "id": 1901744,
      "postDate": "2022-08-16T21:30:17.720Z",
      "content": "<p>The scan <code>1.2.826.0.1.3680043.20574</code> does not include a full cervical spine and should be ignored. Thanks to <a href=\"https://www.kaggle.com/kretes\" target=\"_blank\">@kretes</a> <a href=\"https://www.kaggle.com/code/kretes/pixdim-spacing-distributions?scriptVersionId=102885798#1.2.826.0.1.3680043.20574\" target=\"_blank\">for catching this</a>.</p>",
      "rawMarkdown": "The scan `1.2.826.0.1.3680043.20574` does not include a full cervical spine and should be ignored. Thanks to @kretes [for catching this](https://www.kaggle.com/code/kretes/pixdim-spacing-distributions?scriptVersionId=102885798#1.2.826.0.1.3680043.20574).",
      "votes": 38
    },
    {
      "id": 1914707,
      "postDate": "2022-08-26T10:29:06.163Z",
      "content": "<p>Found another one outlier train sample: <code>1.2.826.0.1.3680043.29952</code>, the patient's neck is tilted by 45 degrees (in sagittal plane), so some scans are obtained not in the axial but in the coronal plane. That can be a problem if model expects all scans/voxel images to be in the same coordinate system. Should be ignored or manually corrected (for example rotated by 45 degrees back).</p>\n<p>180th scan:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F1a091977090a3d41cacfdea89e1ea9b2%2F1.2.826.0.1.3680043.29952.180.png?generation=1661508929832473&amp;alt=media\" alt=\"\"></p>\n<p>3d voxel image visualization, (axial, coronal, sagittal planes from left to right), axial and coronal looks swapped but if you look at lower scan by z axis (inferior-superior), they are swapped back (2nd image):<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F29c5a97e2bd4bea9aec5b7a496cb8dd4%2F1.2.826.0.1.3680043.29952.png?generation=1661509267472097&amp;alt=media\" alt=\"\"><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F5985d8d7a9e3de345c66d70933142ad7%2F1.2.826.0.1.3680043.29952_2.png?generation=1661510193273086&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Found another one outlier train sample: `1.2.826.0.1.3680043.29952`, the patient's neck is tilted by 45 degrees (in sagittal plane), so some scans are obtained not in the axial but in the coronal plane. That can be a problem if model expects all scans/voxel images to be in the same coordinate system. Should be ignored or manually corrected (for example rotated by 45 degrees back).\n\n180th scan:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F1a091977090a3d41cacfdea89e1ea9b2%2F1.2.826.0.1.3680043.29952.180.png?generation=1661508929832473&alt=media)\n\n3d voxel image visualization, (axial, coronal, sagittal planes from left to right), axial and coronal looks swapped but if you look at lower scan by z axis (inferior-superior), they are swapped back (2nd image):\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F29c5a97e2bd4bea9aec5b7a496cb8dd4%2F1.2.826.0.1.3680043.29952.png?generation=1661509267472097&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F5985d8d7a9e3de345c66d70933142ad7%2F1.2.826.0.1.3680043.29952_2.png?generation=1661510193273086&alt=media)",
      "votes": 9
    },
    {
      "id": 1921504,
      "postDate": "2022-08-31T22:07:38.733Z",
      "content": "<p>Also, be careful with the <code>1.2.826.0.1.3680043.20756</code> scan, the slices are duplicated, meaning that there are 2 scans <em>stiched</em> to each other.</p>",
      "rawMarkdown": "Also, be careful with the `1.2.826.0.1.3680043.20756` scan, the slices are duplicated, meaning that there are 2 scans *stiched* to each other.",
      "votes": 4
    },
    {
      "id": 1950376,
      "postDate": "2022-09-22T09:34:56.767Z",
      "content": "<p>I've also found one weird scan, <code>1.2.826.0.1.3680043.8362</code>.  though I'm not sure if this is an outlier.  <br>\n<a href=\"https://www.kaggle.com/code/junhyeonkwon/plot-subjects-outliers\" target=\"_blank\">Here</a> I've made a simple notebook to view the sample.</p>",
      "rawMarkdown": "I've also found one weird scan, ```1.2.826.0.1.3680043.8362```.  though I'm not sure if this is an outlier.  \n[Here](https://www.kaggle.com/code/junhyeonkwon/plot-subjects-outliers) I've made a simple notebook to view the sample.",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 1914707,
      "author_name": "Konstantin Kozlovtsev",
      "author_url": "",
      "post_date": "2022-08-26T10:29:06.163000",
      "content": "<p>Found another one outlier train sample: <code>1.2.826.0.1.3680043.29952</code>, the patient's neck is tilted by 45 degrees (in sagittal plane), so some scans are obtained not in the axial but in the coronal plane. That can be a problem if model expects all scans/voxel images to be in the same coordinate system. Should be ignored or manually corrected (for example rotated by 45 degrees back).</p>\n<p>180th scan:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F1a091977090a3d41cacfdea89e1ea9b2%2F1.2.826.0.1.3680043.29952.180.png?generation=1661508929832473&amp;alt=media\" alt=\"\"></p>\n<p>3d voxel image visualization, (axial, coronal, sagittal planes from left to right), axial and coronal looks swapped but if you look at lower scan by z axis (inferior-superior), they are swapped back (2nd image):<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F29c5a97e2bd4bea9aec5b7a496cb8dd4%2F1.2.826.0.1.3680043.29952.png?generation=1661509267472097&amp;alt=media\" alt=\"\"><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F5985d8d7a9e3de345c66d70933142ad7%2F1.2.826.0.1.3680043.29952_2.png?generation=1661510193273086&amp;alt=media\" alt=\"\"></p>",
      "votes": 9,
      "replies": []
    },
    {
      "id": 1921504,
      "author_name": "Bardia Khosravi",
      "author_url": "",
      "post_date": "2022-08-31T22:07:38.733000",
      "content": "<p>Also, be careful with the <code>1.2.826.0.1.3680043.20756</code> scan, the slices are duplicated, meaning that there are 2 scans <em>stiched</em> to each other.</p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 1950376,
      "author_name": "JunHyeonKwon",
      "author_url": "",
      "post_date": "2022-09-22T09:34:56.767000",
      "content": "<p>I've also found one weird scan, <code>1.2.826.0.1.3680043.8362</code>.  though I'm not sure if this is an outlier.  <br>\n<a href=\"https://www.kaggle.com/code/junhyeonkwon/plot-subjects-outliers\" target=\"_blank\">Here</a> I've made a simple notebook to view the sample.</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1901744": "The scan `1.2.826.0.1.3680043.20574` does not include a full cervical spine and should be ignored. Thanks to @kretes [for catching this](https://www.kaggle.com/code/kretes/pixdim-spacing-distributions?scriptVersionId=102885798#1.2.826.0.1.3680043.20574).",
    "1914707": "Found another one outlier train sample: `1.2.826.0.1.3680043.29952`, the patient's neck is tilted by 45 degrees (in sagittal plane), so some scans are obtained not in the axial but in the coronal plane. That can be a problem if model expects all scans/voxel images to be in the same coordinate system. Should be ignored or manually corrected (for example rotated by 45 degrees back).\n\n180th scan:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F1a091977090a3d41cacfdea89e1ea9b2%2F1.2.826.0.1.3680043.29952.180.png?generation=1661508929832473&alt=media)\n\n3d voxel image visualization, (axial, coronal, sagittal planes from left to right), axial and coronal looks swapped but if you look at lower scan by z axis (inferior-superior), they are swapped back (2nd image):\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F29c5a97e2bd4bea9aec5b7a496cb8dd4%2F1.2.826.0.1.3680043.29952.png?generation=1661509267472097&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F586592%2F5985d8d7a9e3de345c66d70933142ad7%2F1.2.826.0.1.3680043.29952_2.png?generation=1661510193273086&alt=media)",
    "1921504": "Also, be careful with the `1.2.826.0.1.3680043.20756` scan, the slices are duplicated, meaning that there are 2 scans *stiched* to each other.",
    "1950376": "I've also found one weird scan, ```1.2.826.0.1.3680043.8362```.  though I'm not sure if this is an outlier.  \n[Here](https://www.kaggle.com/code/junhyeonkwon/plot-subjects-outliers) I've made a simple notebook to view the sample."
  }
}