{
  "id": 111325,
  "title": "Clarification on Metadata Usage (from the Challenge Organizing Team)",
  "url": "/competitions/rsna-intracranial-hemorrhage-detection/discussion/111325",
  "author_name": "Luciano Prevedello",
  "post_date": "2019-10-04T19:31:00.850000",
  "votes": 53,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Recognizing the confusion generated by the statement “Submission predictions must be based entirely on the pixel data in the provided datasets” and the implications it has on metadata usage, the organizers have decided to retract this rule. The initial intent of the rule was for the algorithm not rely on metadata in order to limit over-fitting and maximize generalizability of the solution based on pixel data only. Given that (1) this generated confusion around metadata usage for preprocessing/model creation capabilities, (2) recognizing the metadata provided in the dataset is de-identified and the available fields do not contain information that can determine if an image contains intracranial hemorrhage, and (3) with the intent not to stifle creativity, the organizing committee has decided to retract this rule and allow all metadata to be used for model creation.</p>\n\n<p>Challenge Organizing Team</p>",
  "messages": [
    {
      "id": 641544,
      "postDate": "2019-10-04T19:31:00.850Z",
      "content": "<p>Recognizing the confusion generated by the statement “Submission predictions must be based entirely on the pixel data in the provided datasets” and the implications it has on metadata usage, the organizers have decided to retract this rule. The initial intent of the rule was for the algorithm not rely on metadata in order to limit over-fitting and maximize generalizability of the solution based on pixel data only. Given that (1) this generated confusion around metadata usage for preprocessing/model creation capabilities, (2) recognizing the metadata provided in the dataset is de-identified and the available fields do not contain information that can determine if an image contains intracranial hemorrhage, and (3) with the intent not to stifle creativity, the organizing committee has decided to retract this rule and allow all metadata to be used for model creation.</p>\n\n<p>Challenge Organizing Team</p>",
      "rawMarkdown": "Recognizing the confusion generated by the statement “Submission predictions must be based entirely on the pixel data in the provided datasets” and the implications it has on metadata usage, the organizers have decided to retract this rule. The initial intent of the rule was for the algorithm not rely on metadata in order to limit over-fitting and maximize generalizability of the solution based on pixel data only. Given that (1) this generated confusion around metadata usage for preprocessing/model creation capabilities, (2) recognizing the metadata provided in the dataset is de-identified and the available fields do not contain information that can determine if an image contains intracranial hemorrhage, and (3) with the intent not to stifle creativity, the organizing committee has decided to retract this rule and allow all metadata to be used for model creation.\n\nChallenge Organizing Team\n",
      "votes": 53
    },
    {
      "id": 641649,
      "postDate": "2019-10-04T23:06:56.833Z",
      "content": "<p><a href=\"/lechuck0\">@lechuck0</a> could you also clarify the 2nd stage of the competition? Will it be from the same machines and institutions? Will there be any systematic change in data distribution (i.e. different mix of patients, different time period, etc)? </p>\n\n<p>Knowing this will allow the competitors to ensure their models do a better job of matching with your goals in the competition.</p>",
      "rawMarkdown": "@lechuck0 could you also clarify the 2nd stage of the competition? Will it be from the same machines and institutions? Will there be any systematic change in data distribution (i.e. different mix of patients, different time period, etc)? \n\nKnowing this will allow the competitors to ensure their models do a better job of matching with your goals in the competition.",
      "votes": 16,
      "replies": [
        {
          "id": 642358,
          "postDate": "2019-10-06T00:28:30.803Z",
          "content": "<p>Stage 2’s dataset quality and time range will match Stage 1. Stage 2 will represent a distinct set of patients though. </p>",
          "rawMarkdown": "Stage 2’s dataset quality and time range will match Stage 1. Stage 2 will represent a distinct set of patients though. ",
          "votes": 6
        }
      ]
    },
    {
      "id": 641641,
      "postDate": "2019-10-04T22:22:42.107Z",
      "content": "<p>Great decision - thanks so much for sorting this out folks! :)  </p>",
      "rawMarkdown": "Great decision - thanks so much for sorting this out folks! :)  ",
      "votes": 10
    },
    {
      "id": 641877,
      "postDate": "2019-10-05T08:59:50.423Z",
      "content": "<p>Thank you for this! This opens up some interesting new avenues using 3D volumes.</p>\n\n<p>For those interested, I made a notebook here: <a href=\"https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata\">https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata</a></p>",
      "rawMarkdown": " Thank you for this! This opens up some interesting new avenues using 3D volumes.\n\nFor those interested, I made a notebook here: https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata",
      "votes": 6
    },
    {
      "id": 644288,
      "postDate": "2019-10-08T15:34:39.887Z",
      "content": "<p>I's say that this is amazing - AGILE from the organization. Respect!\nNow I can finally step in and do some work.\nMulti stage is good anyway.\nregards,\nA</p>",
      "rawMarkdown": "I's say that this is amazing - AGILE from the organization. Respect!\nNow I can finally step in and do some work.\nMulti stage is good anyway.\nregards,\nA",
      "votes": 2
    },
    {
      "id": 641633,
      "postDate": "2019-10-04T21:43:22.463Z",
      "content": "<p>Thoughtful decision from the Organizing Team. Thanks for the clarification. This will increase the modeling possibilities in this challenge.</p>",
      "rawMarkdown": "Thoughtful decision from the Organizing Team. Thanks for the clarification. This will increase the modeling possibilities in this challenge.",
      "votes": 2
    },
    {
      "id": 650043,
      "postDate": "2019-10-16T03:24:23.323Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 641649,
      "author_name": "Jeremy Howard",
      "author_url": "",
      "post_date": "2019-10-04T23:06:56.833000",
      "content": "<p><a href=\"/lechuck0\">@lechuck0</a> could you also clarify the 2nd stage of the competition? Will it be from the same machines and institutions? Will there be any systematic change in data distribution (i.e. different mix of patients, different time period, etc)? </p>\n\n<p>Knowing this will allow the competitors to ensure their models do a better job of matching with your goals in the competition.</p>",
      "votes": 16,
      "replies": [
        {
          "id": 642358,
          "author_name": "Luciano Prevedello",
          "author_url": "",
          "post_date": "2019-10-06T00:28:30.803000",
          "content": "<p>Stage 2’s dataset quality and time range will match Stage 1. Stage 2 will represent a distinct set of patients though. </p>",
          "votes": 6,
          "replies": []
        }
      ]
    },
    {
      "id": 641641,
      "author_name": "Jeremy Howard",
      "author_url": "",
      "post_date": "2019-10-04T22:22:42.107000",
      "content": "<p>Great decision - thanks so much for sorting this out folks! :)  </p>",
      "votes": 10,
      "replies": []
    },
    {
      "id": 641877,
      "author_name": "datasaurus",
      "author_url": "",
      "post_date": "2019-10-05T08:59:50.423000",
      "content": "<p>Thank you for this! This opens up some interesting new avenues using 3D volumes.</p>\n\n<p>For those interested, I made a notebook here: <a href=\"https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata\">https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata</a></p>",
      "votes": 6,
      "replies": []
    },
    {
      "id": 644288,
      "author_name": "Adrian Zinovei",
      "author_url": "",
      "post_date": "2019-10-08T15:34:39.887000",
      "content": "<p>I's say that this is amazing - AGILE from the organization. Respect!\nNow I can finally step in and do some work.\nMulti stage is good anyway.\nregards,\nA</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 641633,
      "author_name": "FelipeKitamura, MD, PhD",
      "author_url": "",
      "post_date": "2019-10-04T21:43:22.463000",
      "content": "<p>Thoughtful decision from the Organizing Team. Thanks for the clarification. This will increase the modeling possibilities in this challenge.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 650043,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-10-16T03:24:23.323000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "641544": "Recognizing the confusion generated by the statement “Submission predictions must be based entirely on the pixel data in the provided datasets” and the implications it has on metadata usage, the organizers have decided to retract this rule. The initial intent of the rule was for the algorithm not rely on metadata in order to limit over-fitting and maximize generalizability of the solution based on pixel data only. Given that (1) this generated confusion around metadata usage for preprocessing/model creation capabilities, (2) recognizing the metadata provided in the dataset is de-identified and the available fields do not contain information that can determine if an image contains intracranial hemorrhage, and (3) with the intent not to stifle creativity, the organizing committee has decided to retract this rule and allow all metadata to be used for model creation.\n\nChallenge Organizing Team\n",
    "641649": "@lechuck0 could you also clarify the 2nd stage of the competition? Will it be from the same machines and institutions? Will there be any systematic change in data distribution (i.e. different mix of patients, different time period, etc)? \n\nKnowing this will allow the competitors to ensure their models do a better job of matching with your goals in the competition.",
    "641641": "Great decision - thanks so much for sorting this out folks! :)  ",
    "641877": " Thank you for this! This opens up some interesting new avenues using 3D volumes.\n\nFor those interested, I made a notebook here: https://www.kaggle.com/anjum48/reconstructing-3d-volumes-from-metadata",
    "644288": "I's say that this is amazing - AGILE from the organization. Respect!\nNow I can finally step in and do some work.\nMulti stage is good anyway.\nregards,\nA",
    "641633": "Thoughtful decision from the Organizing Team. Thanks for the clarification. This will increase the modeling possibilities in this challenge.",
    "650043": ""
  }
}