{
  "id": 110089,
  "title": "patient wise versus image wise submissions ?",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/110089",
  "author_name": "simjeg",
  "post_date": "2019-09-24T16:23:24.264000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello Hello,</p>\n\n<p>There are 600k images in the training set, corresponding to around 17k patients.\nUsing the ID of the patient could be useful to regularize image wise predictions. For instance there is a low probability to detect the 5 different classes for the same patient. It could also be used to work in 3D rather than 2D.</p>\n\n<p>Are we allowed to take patient wise information into account in our predictions ?\nThanks,\nSimon</p>",
  "messages": [
    {
      "id": 633281,
      "postDate": "2019-09-24T17:08:12.297Z",
      "content": "<p>Hi <a href=\"/simjeg\">@simjeg</a> \nMaybe this thread could partly answer your question: <a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646</a>\nEspecially Phil Cullington's answer:</p>\n\n<p>Good questions!</p>\n\n<p>kambarakun - it is not against the rules to create a 3D construction using the series ID.</p>\n\n<p>lopuhin - sorry for the ambiguity. We're approaching this on a case-by-case basis. If you have a use for the metadata that falls outside of what has already been allowed, please ask, and the host will clarify. All of the provided fields were included so they could be used, but the exact use should be cleared to make sure you're falling within the guidelines of the rules.</p>",
      "rawMarkdown": "Hi @simjeg \nMaybe this thread could partly answer your question: https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646\nEspecially Phil Cullington's answer:\n\nGood questions!\n\nkambarakun - it is not against the rules to create a 3D construction using the series ID.\n\nlopuhin - sorry for the ambiguity. We're approaching this on a case-by-case basis. If you have a use for the metadata that falls outside of what has already been allowed, please ask, and the host will clarify. All of the provided fields were included so they could be used, but the exact use should be cleared to make sure you're falling within the guidelines of the rules.",
      "votes": 1
    },
    {
      "id": 633263,
      "postDate": "2019-09-24T16:23:24.263Z",
      "content": "<p>Hello Hello,</p>\n\n<p>There are 600k images in the training set, corresponding to around 17k patients.\nUsing the ID of the patient could be useful to regularize image wise predictions. For instance there is a low probability to detect the 5 different classes for the same patient. It could also be used to work in 3D rather than 2D.</p>\n\n<p>Are we allowed to take patient wise information into account in our predictions ?\nThanks,\nSimon</p>",
      "rawMarkdown": "Hello Hello,\n\nThere are 600k images in the training set, corresponding to around 17k patients.\nUsing the ID of the patient could be useful to regularize image wise predictions. For instance there is a low probability to detect the 5 different classes for the same patient. It could also be used to work in 3D rather than 2D.\n\nAre we allowed to take patient wise information into account in our predictions ?\nThanks,\nSimon",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 633281,
      "author_name": "Benoît Koenig",
      "author_url": "",
      "post_date": "2019-09-24T17:08:12.297000",
      "content": "<p>Hi <a href=\"/simjeg\">@simjeg</a> \nMaybe this thread could partly answer your question: <a href=\"https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646\">https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646</a>\nEspecially Phil Cullington's answer:</p>\n\n<p>Good questions!</p>\n\n<p>kambarakun - it is not against the rules to create a 3D construction using the series ID.</p>\n\n<p>lopuhin - sorry for the ambiguity. We're approaching this on a case-by-case basis. If you have a use for the metadata that falls outside of what has already been allowed, please ask, and the host will clarify. All of the provided fields were included so they could be used, but the exact use should be cleared to make sure you're falling within the guidelines of the rules.</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "633281": "Hi @simjeg \nMaybe this thread could partly answer your question: https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/discussion/109479#latest-632646\nEspecially Phil Cullington's answer:\n\nGood questions!\n\nkambarakun - it is not against the rules to create a 3D construction using the series ID.\n\nlopuhin - sorry for the ambiguity. We're approaching this on a case-by-case basis. If you have a use for the metadata that falls outside of what has already been allowed, please ask, and the host will clarify. All of the provided fields were included so they could be used, but the exact use should be cleared to make sure you're falling within the guidelines of the rules.",
    "633263": "Hello Hello,\n\nThere are 600k images in the training set, corresponding to around 17k patients.\nUsing the ID of the patient could be useful to regularize image wise predictions. For instance there is a low probability to detect the 5 different classes for the same patient. It could also be used to work in 3D rather than 2D.\n\nAre we allowed to take patient wise information into account in our predictions ?\nThanks,\nSimon"
  }
}