{
  "id": 454353,
  "title": "Basic questions about resolution and sparsely segmentation",
  "url": "/competitions/blood-vessel-segmentation/discussion/454353",
  "author_name": "",
  "post_date": "2023-11-09T21:47:24.024675300Z",
  "votes": 24,
  "comment_count": 2,
  "views": 0,
  "content": "<ol>\n<li><p>\"50um resolution.\" - Does this mean 1 voxel on the image corresponds to 50x50x50um physical distance?</p></li>\n<li><p>\"Uses beamline BM05.\" Is this more of a physical instrument used in manual segmentation (so the segmentation is manually done by people) or it involve some sort of algorithm that does the segmentation auto/semi-automatically?</p></li>\n<li><p>\"Sparsely segmented (about 65%).\" What does this mean exactly? My understanding is - probably due to the labor-intensive nature of the manual labeling, only some of the vasculature was labeled, but if not all was segmented, how does one know the labeled ones is around 65%? Or is this value calculated based on area - e.g. 65% of the image volume has been annotated, or it means something else?</p></li>\n<li><p>Is there any bias in the sparsely segmented case? e.g. the 65% labeled region is the easier ones vs the rest is harder to annotate? </p></li>\n<li><p>\"The entire 3D arterial vascular tree has been densely segmented\" Does it mean it's segmented 100% i.e. there are no missing pieces that did not get annotated?</p></li>\n</ol>",
  "messages": [
    {
      "id": "2519288",
      "postDate": "11/09/2023 21:47:24",
      "content": "<ol>\n<li><p>\"50um resolution.\" - Does this mean 1 voxel on the image corresponds to 50x50x50um physical distance?</p></li>\n<li><p>\"Uses beamline BM05.\" Is this more of a physical instrument used in manual segmentation (so the segmentation is manually done by people) or it involve some sort of algorithm that does the segmentation auto/semi-automatically?</p></li>\n<li><p>\"Sparsely segmented (about 65%).\" What does this mean exactly? My understanding is - probably due to the labor-intensive nature of the manual labeling, only some of the vasculature was labeled, but if not all was segmented, how does one know the labeled ones is around 65%? Or is this value calculated based on area - e.g. 65% of the image volume has been annotated, or it means something else?</p></li>\n<li><p>Is there any bias in the sparsely segmented case? e.g. the 65% labeled region is the easier ones vs the rest is harder to annotate? </p></li>\n<li><p>\"The entire 3D arterial vascular tree has been densely segmented\" Does it mean it's segmented 100% i.e. there are no missing pieces that did not get annotated?</p></li>\n</ol>",
      "rawMarkdown": "1. \"50um resolution.\" - Does this mean 1 voxel on the image corresponds to 50x50x50um physical distance?\n\n2. \"Uses beamline BM05.\" Is this more of a physical instrument used in manual segmentation (so the segmentation is manually done by people) or it involve some sort of algorithm that does the segmentation auto/semi-automatically?\n\n3. \"Sparsely segmented (about 65%).\" What does this mean exactly? My understanding is - probably due to the labor-intensive nature of the manual labeling, only some of the vasculature was labeled, but if not all was segmented, how does one know the labeled ones is around 65%? Or is this value calculated based on area - e.g. 65% of the image volume has been annotated, or it means something else?\n\n4. Is there any bias in the sparsely segmented case? e.g. the 65% labeled region is the easier ones vs the rest is harder to annotate? \n\n5. \"The entire 3D arterial vascular tree has been densely segmented\" Does it mean it's segmented 100% i.e. there are no missing pieces that did not get annotated?",
      "votes": null
    },
    {
      "id": "2520026",
      "postDate": "11/10/2023 14:34:59",
      "content": "<p>Hi so, here are some answers that will hopefully help: </p>\n<ol>\n<li>Yes</li>\n<li>BM05 is a beamline at the synchrotron where the data are collected, (nothing to do with segmentation).</li>\n<li>Yes it is manually segmented by annotators. We used a two annotator and validation strategy. After the first annotation by an experienced annotator is done, a second, independent, experienced annotator meticulously conducts 3D proofreading of the binary labels filling, in any missing vessel labels in the three orthogonal planes. After this, five randomly selected 2D circular regions of the image volume are selected, a third annotator marks all vessels visible in the 2D region (1 mark per vessel) then, using the labels, counts percentage of vessels that were missed i.e. the FN. This gives our % segmentation that we provide. For the complete segmentations, we just keep going around this loop iteratively to capture all vessels.  </li>\n</ol>",
      "rawMarkdown": "Hi so, here are some answers that will hopefully help: \n1. Yes\n2. BM05 is a beamline at the synchrotron where the data are collected, (nothing to do with segmentation).\n3. Yes it is manually segmented by annotators. We used a two annotator and validation strategy. After the first annotation by an experienced annotator is done, a second, independent, experienced annotator meticulously conducts 3D proofreading of the binary labels filling, in any missing vessel labels in the three orthogonal planes. After this, five randomly selected 2D circular regions of the image volume are selected, a third annotator marks all vessels visible in the 2D region (1 mark per vessel) then, using the labels, counts percentage of vessels that were missed i.e. the FN. This gives our % segmentation that we provide. For the complete segmentations, we just keep going around this loop iteratively to capture all vessels.",
      "votes": null
    },
    {
      "id": "2520072",
      "postDate": "11/10/2023 15:01:01",
      "content": "<p><a href=\"https://www.kaggle.com/clairewalsh\" target=\"_blank\">@clairewalsh</a> <br>\nThank you for your detailed explanation. Is the full test set densely segmented?</p>",
      "rawMarkdown": "clairewalsh \nThank you for your detailed explanation. Is the full test set densely segmented?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2520026,
      "author_name": "clairewalsh",
      "author_url": "",
      "post_date": "11/10/2023 14:34:59",
      "content": "<p>Hi so, here are some answers that will hopefully help: </p>\n<ol>\n<li>Yes</li>\n<li>BM05 is a beamline at the synchrotron where the data are collected, (nothing to do with segmentation).</li>\n<li>Yes it is manually segmented by annotators. We used a two annotator and validation strategy. After the first annotation by an experienced annotator is done, a second, independent, experienced annotator meticulously conducts 3D proofreading of the binary labels filling, in any missing vessel labels in the three orthogonal planes. After this, five randomly selected 2D circular regions of the image volume are selected, a third annotator marks all vessels visible in the 2D region (1 mark per vessel) then, using the labels, counts percentage of vessels that were missed i.e. the FN. This gives our % segmentation that we provide. For the complete segmentations, we just keep going around this loop iteratively to capture all vessels.  </li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 2520072,
          "author_name": "amanatsu",
          "author_url": "",
          "post_date": "11/10/2023 15:01:01",
          "content": "<p><a href=\"https://www.kaggle.com/clairewalsh\" target=\"_blank\">@clairewalsh</a> <br>\nThank you for your detailed explanation. Is the full test set densely segmented?</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2519288": "1. \"50um resolution.\" - Does this mean 1 voxel on the image corresponds to 50x50x50um physical distance?\n\n2. \"Uses beamline BM05.\" Is this more of a physical instrument used in manual segmentation (so the segmentation is manually done by people) or it involve some sort of algorithm that does the segmentation auto/semi-automatically?\n\n3. \"Sparsely segmented (about 65%).\" What does this mean exactly? My understanding is - probably due to the labor-intensive nature of the manual labeling, only some of the vasculature was labeled, but if not all was segmented, how does one know the labeled ones is around 65%? Or is this value calculated based on area - e.g. 65% of the image volume has been annotated, or it means something else?\n\n4. Is there any bias in the sparsely segmented case? e.g. the 65% labeled region is the easier ones vs the rest is harder to annotate? \n\n5. \"The entire 3D arterial vascular tree has been densely segmented\" Does it mean it's segmented 100% i.e. there are no missing pieces that did not get annotated?",
    "2520026": "Hi so, here are some answers that will hopefully help: \n1. Yes\n2. BM05 is a beamline at the synchrotron where the data are collected, (nothing to do with segmentation).\n3. Yes it is manually segmented by annotators. We used a two annotator and validation strategy. After the first annotation by an experienced annotator is done, a second, independent, experienced annotator meticulously conducts 3D proofreading of the binary labels filling, in any missing vessel labels in the three orthogonal planes. After this, five randomly selected 2D circular regions of the image volume are selected, a third annotator marks all vessels visible in the 2D region (1 mark per vessel) then, using the labels, counts percentage of vessels that were missed i.e. the FN. This gives our % segmentation that we provide. For the complete segmentations, we just keep going around this loop iteratively to capture all vessels.",
    "2520072": "clairewalsh \nThank you for your detailed explanation. Is the full test set densely segmented?"
  },
  "source": "meta"
}