{
  "id": 19079,
  "title": "Re-slicing",
  "url": "/competitions/second-annual-data-science-bowl/discussion/19079",
  "author_name": "",
  "post_date": "2016-02-19T16:29:38.540Z",
  "votes": null,
  "comment_count": 4,
  "views": 818,
  "content": "<p>Many of the tutorals talk about rescaling and rotating. However, one could probably also re-slice DICOMs to align the 3D orientation across stacks. I am new to medical imaging, so please interpret in this context: Does it make sense to reslice and align patient orientation? Or would you inherently lose a lot of data and/or this is not the right way to think about it? </p>",
  "messages": [
    {
      "id": "108755",
      "postDate": "02/19/2016 16:29:38",
      "content": "<p>Many of the tutorals talk about rescaling and rotating. However, one could probably also re-slice DICOMs to align the 3D orientation across stacks. I am new to medical imaging, so please interpret in this context: Does it make sense to reslice and align patient orientation? Or would you inherently lose a lot of data and/or this is not the right way to think about it? </p>",
      "rawMarkdown": "Many of the tutorals talk about rescaling and rotating. However, one could probably also re-slice DICOMs to align the 3D orientation across stacks. I am new to medical imaging, so please interpret in this context: Does it make sense to reslice and align patient orientation? Or would you inherently lose a lot of data and/or this is not the right way to think about it?",
      "votes": null
    },
    {
      "id": "108768",
      "postDate": "02/19/2016 18:12:54",
      "content": "<p>In my point of view, it's depends what will do with the data in the next steps.</p>\n\n<p>You can lose by one side, and win by another.</p>",
      "rawMarkdown": "In my point of view, it's depends what will do with the data in the next steps.\r\n\r\nYou can lose by one side, and win by another.",
      "votes": null
    },
    {
      "id": "108789",
      "postDate": "02/19/2016 21:42:57",
      "content": "<p>@thanks alvaro. in your best guess, considering that this is MRI data, would you expect that the data loss would be great? (that would be my suspicion - but I am not sure)</p>",
      "rawMarkdown": "thanks alvaro. in your best guess, considering that this is MRI data, would you expect that the data loss would be great? (that would be my suspicion - but I am not sure)",
      "votes": null
    },
    {
      "id": "108792",
      "postDate": "02/19/2016 21:58:34",
      "content": "<p>In my opinion it depends of how much money you have to spend with calculations (energy or a virtual machine for a provider) and your <strong>mathematical background</strong>.</p>\n\n<p>Have you been thinking  of how much power of calculation you have ?</p>\n\n<p>I see it has a important variable to determine the better strategy.</p>\n\n<p>In my opinion, with better calculation power, less you need to work because the computer can use better frameworks and GPUs to learn about the data.</p>\n\n<p>However, with less calculation power more you need to spend with a better mathematical background, because you need to handle the complexity.</p>\n\n<p>Another point, i think it's important have a <strong>systemic vision</strong>, slice the image can be a good idea if you need to standardize the data entry.</p>\n\n<p>It's a point of view.</p>",
      "rawMarkdown": "In my opinion it depends of how much money you have to spend with calculations (energy or a virtual machine for a provider) and your **mathematical background**.\r\n\r\nHave you been thinking  of how much power of calculation you have ?\r\n\r\nI see it has a important variable to determine the better strategy.\r\n\r\nIn my opinion, with better calculation power, less you need to work because the computer can use better frameworks and GPUs to learn about the data.\r\n\r\nHowever, with less calculation power more you need to spend with a better mathematical background, because you need to handle the complexity.\r\n\r\nAnother point, i think it's important have a **systemic vision**, slice the image can be a good idea if you need to standardize the data entry.\r\n\r\nIt's a point of view.",
      "votes": null
    },
    {
      "id": "108817",
      "postDate": "02/20/2016 07:08:46",
      "content": "<p>Hi WD,</p>\n\n<p>My system vision is:\n1) find the left ventricle of the heart for each patient - for this step I append slices by time within each sax location\n2) crop the image around the left ventricle - this is done one image at a time\n3) binarise and find the boundary of the left ventricle chamber within each image - this is done one image at a time\n4) find the time period with diastole and the time period with systole - for this I compare slices by time period within each sax location\n5) calculate the volume by creating a 3D data of the chamber - for this step I append slices by sax location for a particular time</p>\n\n<p>My leaderboard score doesn't reflect this approach yet, as it is a lot of work to implement all of these steps.</p>\n\n<p>My choice of approach was influenced by my computer resources (64GB of RAM and a good GPU), and my time resources (most of the work is done in my spare time), and my background (I already know how to do image morphology and image processing).</p>\n\n<p>Colin</p>",
      "rawMarkdown": "Hi WD,\r\n\r\nMy system vision is:\r\n1) find the left ventricle of the heart for each patient - for this step I append slices by time within each sax location\r\n2) crop the image around the left ventricle - this is done one image at a time\r\n3) binarise and find the boundary of the left ventricle chamber within each image - this is done one image at a time\r\n4) find the time period with diastole and the time period with systole - for this I compare slices by time period within each sax location\r\n5) calculate the volume by creating a 3D data of the chamber - for this step I append slices by sax location for a particular time\r\n\r\nMy leaderboard score doesn't reflect this approach yet, as it is a lot of work to implement all of these steps.\r\n\r\nMy choice of approach was influenced by my computer resources (64GB of RAM and a good GPU), and my time resources (most of the work is done in my spare time), and my background (I already know how to do image morphology and image processing).\r\n\r\nColin",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 108768,
      "author_name": "alvaroosvaldo",
      "author_url": "",
      "post_date": "02/19/2016 18:12:54",
      "content": "<p>In my point of view, it's depends what will do with the data in the next steps.</p>\n\n<p>You can lose by one side, and win by another.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 108789,
      "author_name": "wouterd1",
      "author_url": "",
      "post_date": "02/19/2016 21:42:57",
      "content": "<p>@thanks alvaro. in your best guess, considering that this is MRI data, would you expect that the data loss would be great? (that would be my suspicion - but I am not sure)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 108792,
      "author_name": "alvaroosvaldo",
      "author_url": "",
      "post_date": "02/19/2016 21:58:34",
      "content": "<p>In my opinion it depends of how much money you have to spend with calculations (energy or a virtual machine for a provider) and your <strong>mathematical background</strong>.</p>\n\n<p>Have you been thinking  of how much power of calculation you have ?</p>\n\n<p>I see it has a important variable to determine the better strategy.</p>\n\n<p>In my opinion, with better calculation power, less you need to work because the computer can use better frameworks and GPUs to learn about the data.</p>\n\n<p>However, with less calculation power more you need to spend with a better mathematical background, because you need to handle the complexity.</p>\n\n<p>Another point, i think it's important have a <strong>systemic vision</strong>, slice the image can be a good idea if you need to standardize the data entry.</p>\n\n<p>It's a point of view.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 108817,
      "author_name": "colinpriest",
      "author_url": "",
      "post_date": "02/20/2016 07:08:46",
      "content": "<p>Hi WD,</p>\n\n<p>My system vision is:\n1) find the left ventricle of the heart for each patient - for this step I append slices by time within each sax location\n2) crop the image around the left ventricle - this is done one image at a time\n3) binarise and find the boundary of the left ventricle chamber within each image - this is done one image at a time\n4) find the time period with diastole and the time period with systole - for this I compare slices by time period within each sax location\n5) calculate the volume by creating a 3D data of the chamber - for this step I append slices by sax location for a particular time</p>\n\n<p>My leaderboard score doesn't reflect this approach yet, as it is a lot of work to implement all of these steps.</p>\n\n<p>My choice of approach was influenced by my computer resources (64GB of RAM and a good GPU), and my time resources (most of the work is done in my spare time), and my background (I already know how to do image morphology and image processing).</p>\n\n<p>Colin</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "108755": "Many of the tutorals talk about rescaling and rotating. However, one could probably also re-slice DICOMs to align the 3D orientation across stacks. I am new to medical imaging, so please interpret in this context: Does it make sense to reslice and align patient orientation? Or would you inherently lose a lot of data and/or this is not the right way to think about it?",
    "108768": "In my point of view, it's depends what will do with the data in the next steps.\r\n\r\nYou can lose by one side, and win by another.",
    "108789": "thanks alvaro. in your best guess, considering that this is MRI data, would you expect that the data loss would be great? (that would be my suspicion - but I am not sure)",
    "108792": "In my opinion it depends of how much money you have to spend with calculations (energy or a virtual machine for a provider) and your **mathematical background**.\r\n\r\nHave you been thinking  of how much power of calculation you have ?\r\n\r\nI see it has a important variable to determine the better strategy.\r\n\r\nIn my opinion, with better calculation power, less you need to work because the computer can use better frameworks and GPUs to learn about the data.\r\n\r\nHowever, with less calculation power more you need to spend with a better mathematical background, because you need to handle the complexity.\r\n\r\nAnother point, i think it's important have a **systemic vision**, slice the image can be a good idea if you need to standardize the data entry.\r\n\r\nIt's a point of view.",
    "108817": "Hi WD,\r\n\r\nMy system vision is:\r\n1) find the left ventricle of the heart for each patient - for this step I append slices by time within each sax location\r\n2) crop the image around the left ventricle - this is done one image at a time\r\n3) binarise and find the boundary of the left ventricle chamber within each image - this is done one image at a time\r\n4) find the time period with diastole and the time period with systole - for this I compare slices by time period within each sax location\r\n5) calculate the volume by creating a 3D data of the chamber - for this step I append slices by sax location for a particular time\r\n\r\nMy leaderboard score doesn't reflect this approach yet, as it is a lot of work to implement all of these steps.\r\n\r\nMy choice of approach was influenced by my computer resources (64GB of RAM and a good GPU), and my time resources (most of the work is done in my spare time), and my background (I already know how to do image morphology and image processing).\r\n\r\nColin"
  },
  "source": "meta"
}