{
  "id": 596852,
  "title": "Can we further process the provided coordinates?",
  "url": "/competitions/rsna-intracranial-aneurysm-detection/discussion/596852",
  "author_name": "Chia Chuan",
  "post_date": "2025-08-05T14:06:44.430000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>For each aneurysm, the competition provides the corresponding image (SOPInstanceUID) along with the coordinates near its center. We are considering whether we can leverage these coordinates to enhance our training data and improve annotation efficiency by generating additional human-verified labels from the provided coordinates, making better use of the provided information.</p>\n<p>Our potential processing methods include:</p>\n<ol>\n<li>Manual annotation</li>\n<li>Assisted annotation using algorithms or AI models</li>\n</ol>\n<p>Regarding AI-assisted annotation:<br>\nWe plan to use either public models or models trained on private data to generate initial predictions. All predictions will be manually reviewed and corrected before being used as training data. We will not directly use a pre-trained model to predict the test set. We will not use models trained with private data as pre-trained weights.<br>\nWould this annotation and training strategy be allowed under the rules of this competition?<br>\nAdditionally:<br>\nIs the manual review and correction step necessary for this workflow to remain compliant?<br>\nCan the same strategy also be applied to vessel-related annotations to enrich our training data?</p>",
  "messages": [
    {
      "id": 3265692,
      "postDate": "2025-08-07T20:36:29.807Z",
      "content": "<p>I think this is all fine. Just be aware that neither coordinates nor vessel segmentations are provided for the test set. </p>",
      "rawMarkdown": "I think this is all fine. Just be aware that neither coordinates nor vessel segmentations are provided for the test set. ",
      "votes": 1
    },
    {
      "id": 3263654,
      "postDate": "2025-08-05T14:06:44.430Z",
      "content": "<p>For each aneurysm, the competition provides the corresponding image (SOPInstanceUID) along with the coordinates near its center. We are considering whether we can leverage these coordinates to enhance our training data and improve annotation efficiency by generating additional human-verified labels from the provided coordinates, making better use of the provided information.</p>\n<p>Our potential processing methods include:</p>\n<ol>\n<li>Manual annotation</li>\n<li>Assisted annotation using algorithms or AI models</li>\n</ol>\n<p>Regarding AI-assisted annotation:<br>\nWe plan to use either public models or models trained on private data to generate initial predictions. All predictions will be manually reviewed and corrected before being used as training data. We will not directly use a pre-trained model to predict the test set. We will not use models trained with private data as pre-trained weights.<br>\nWould this annotation and training strategy be allowed under the rules of this competition?<br>\nAdditionally:<br>\nIs the manual review and correction step necessary for this workflow to remain compliant?<br>\nCan the same strategy also be applied to vessel-related annotations to enrich our training data?</p>",
      "rawMarkdown": "For each aneurysm, the competition provides the corresponding image (SOPInstanceUID) along with the coordinates near its center. We are considering whether we can leverage these coordinates to enhance our training data and improve annotation efficiency by generating additional human-verified labels from the provided coordinates, making better use of the provided information.\n\nOur potential processing methods include:\n1. Manual annotation\n2. Assisted annotation using algorithms or AI models\n\nRegarding AI-assisted annotation:\nWe plan to use either public models or models trained on private data to generate initial predictions. All predictions will be manually reviewed and corrected before being used as training data. We will not directly use a pre-trained model to predict the test set. We will not use models trained with private data as pre-trained weights.\nWould this annotation and training strategy be allowed under the rules of this competition?\nAdditionally:\nIs the manual review and correction step necessary for this workflow to remain compliant?\nCan the same strategy also be applied to vessel-related annotations to enrich our training data?\n\n",
      "votes": 2
    }
  ],
  "comments": [
    {
      "id": 3265692,
      "author_name": "Evan Calabrese",
      "author_url": "",
      "post_date": "2025-08-07T20:36:29.807000",
      "content": "<p>I think this is all fine. Just be aware that neither coordinates nor vessel segmentations are provided for the test set. </p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3265692": "I think this is all fine. Just be aware that neither coordinates nor vessel segmentations are provided for the test set. ",
    "3263654": "For each aneurysm, the competition provides the corresponding image (SOPInstanceUID) along with the coordinates near its center. We are considering whether we can leverage these coordinates to enhance our training data and improve annotation efficiency by generating additional human-verified labels from the provided coordinates, making better use of the provided information.\n\nOur potential processing methods include:\n1. Manual annotation\n2. Assisted annotation using algorithms or AI models\n\nRegarding AI-assisted annotation:\nWe plan to use either public models or models trained on private data to generate initial predictions. All predictions will be manually reviewed and corrected before being used as training data. We will not directly use a pre-trained model to predict the test set. We will not use models trained with private data as pre-trained weights.\nWould this annotation and training strategy be allowed under the rules of this competition?\nAdditionally:\nIs the manual review and correction step necessary for this workflow to remain compliant?\nCan the same strategy also be applied to vessel-related annotations to enrich our training data?\n\n"
  }
}