{
  "id": 266685,
  "title": "Is every series in train, validation and test of the same anatomy coordinate system ? (LPS, RAI, ..) ?",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/266685",
  "author_name": "",
  "post_date": "2021-08-20T02:12:31.139102900Z",
  "votes": 9,
  "comment_count": 3,
  "views": 0,
  "content": "<p>It is very important that we know the all the data in train, validation and test set is of the same anatomy coordinate system (ACS). Because even if we have the same ImageOrientationPatient matrix, it could end up different orientation transformations in different ACS. </p>\n<p><img src=\"http://www.grahamwideman.com/gw/brain/orientation/rlapsi.gif\" alt=\"ACS demonstration\"></p>\n<p>Some public notebooks propose ways to resample all the patient voxels to a fixed reference. But it only makes sense if all the patient voxels are in the same ACS.</p>\n<p>Can you confirm this? <a href=\"https://www.kaggle.com/ujjwalbaid\" target=\"_blank\">@ujjwalbaid</a>. Thank you!</p>",
  "messages": [
    {
      "id": "1482285",
      "postDate": "08/20/2021 02:12:31",
      "content": "<p>It is very important that we know the all the data in train, validation and test set is of the same anatomy coordinate system (ACS). Because even if we have the same ImageOrientationPatient matrix, it could end up different orientation transformations in different ACS. </p>\n<p><img src=\"http://www.grahamwideman.com/gw/brain/orientation/rlapsi.gif\" alt=\"ACS demonstration\"></p>\n<p>Some public notebooks propose ways to resample all the patient voxels to a fixed reference. But it only makes sense if all the patient voxels are in the same ACS.</p>\n<p>Can you confirm this? <a href=\"https://www.kaggle.com/ujjwalbaid\" target=\"_blank\">@ujjwalbaid</a>. Thank you!</p>",
      "rawMarkdown": "It is very important that we know the all the data in train, validation and test set is of the same anatomy coordinate system (ACS). Because even if we have the same ImageOrientationPatient matrix, it could end up different orientation transformations in different ACS. \n\n![ACS demonstration](http://www.grahamwideman.com/gw/brain/orientation/rlapsi.gif)\n\nSome public notebooks propose ways to resample all the patient voxels to a fixed reference. But it only makes sense if all the patient voxels are in the same ACS.\n\nCan you confirm this? @ujjwalbaid. Thank you!",
      "votes": null
    },
    {
      "id": "1482366",
      "postDate": "08/20/2021 03:40:30",
      "content": "<p>I can confirm this, upvoting!</p>",
      "rawMarkdown": "I can confirm this, upvoting!",
      "votes": null
    },
    {
      "id": "1482802",
      "postDate": "08/20/2021 09:39:23",
      "content": "<p>Will you explain how do can you confirm this ?</p>",
      "rawMarkdown": "Will you explain how do can you confirm this ?",
      "votes": null
    },
    {
      "id": "1483125",
      "postDate": "08/20/2021 13:06:48",
      "content": "<p>Determining the spatial relationship between images/series requires the DICOM tag FrameOfReferenceUID (0020,0052). </p>\n<p>From the DICOM standard -&gt; <a href=\"http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html\" target=\"_blank\">http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html</a></p>\n<blockquote>\n  <p>All images in a series that share the same Frame of Reference UID shall be spatially related to each other.</p>\n</blockquote>\n<p>I checked a few files and did not find any FrameOfReference tags. There may be some studies with this tag, but my overall assessment is that we cannot assume these series exist in the same space.</p>",
      "rawMarkdown": "Determining the spatial relationship between images/series requires the DICOM tag FrameOfReferenceUID (0020,0052). \n\nFrom the DICOM standard -> http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html\n\n> All images in a series that share the same Frame of Reference UID shall be spatially related to each other.\n\nI checked a few files and did not find any FrameOfReference tags. There may be some studies with this tag, but my overall assessment is that we cannot assume these series exist in the same space.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1482366,
      "author_name": "rishirajacharya",
      "author_url": "",
      "post_date": "08/20/2021 03:40:30",
      "content": "<p>I can confirm this, upvoting!</p>",
      "votes": null,
      "replies": [
        {
          "id": 1482802,
          "author_name": "namgalielei",
          "author_url": "",
          "post_date": "08/20/2021 09:39:23",
          "content": "<p>Will you explain how do can you confirm this ?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1483125,
      "author_name": "davidbroberts",
      "author_url": "",
      "post_date": "08/20/2021 13:06:48",
      "content": "<p>Determining the spatial relationship between images/series requires the DICOM tag FrameOfReferenceUID (0020,0052). </p>\n<p>From the DICOM standard -&gt; <a href=\"http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html\" target=\"_blank\">http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html</a></p>\n<blockquote>\n  <p>All images in a series that share the same Frame of Reference UID shall be spatially related to each other.</p>\n</blockquote>\n<p>I checked a few files and did not find any FrameOfReference tags. There may be some studies with this tag, but my overall assessment is that we cannot assume these series exist in the same space.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1482285": "It is very important that we know the all the data in train, validation and test set is of the same anatomy coordinate system (ACS). Because even if we have the same ImageOrientationPatient matrix, it could end up different orientation transformations in different ACS. \n\n![ACS demonstration](http://www.grahamwideman.com/gw/brain/orientation/rlapsi.gif)\n\nSome public notebooks propose ways to resample all the patient voxels to a fixed reference. But it only makes sense if all the patient voxels are in the same ACS.\n\nCan you confirm this? @ujjwalbaid. Thank you!",
    "1482366": "I can confirm this, upvoting!",
    "1482802": "Will you explain how do can you confirm this ?",
    "1483125": "Determining the spatial relationship between images/series requires the DICOM tag FrameOfReferenceUID (0020,0052). \n\nFrom the DICOM standard -> http://dicom.nema.org/medical/Dicom/2016b/output/chtml/part03/sect_C.7.4.html\n\n> All images in a series that share the same Frame of Reference UID shall be spatially related to each other.\n\nI checked a few files and did not find any FrameOfReference tags. There may be some studies with this tag, but my overall assessment is that we cannot assume these series exist in the same space."
  },
  "source": "meta"
}