{
  "id": 463372,
  "title": "Why do we analyze inaccurate images using CS, rather than reconstruct accurate images using CS?",
  "url": "/competitions/blood-vessel-segmentation/discussion/463372",
  "author_name": "",
  "post_date": "2023-12-24T23:20:26.869683700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>As I understand, imaging methods basically throw something (like x-rays) at an object (like a kidney), then collect some signal resulting from that throwing (like received intensities of x-rays), from which they then try to reconstruct a 3-d image of the object.</p>\n<p>It seems natural to ask a question, why isn't it better to improve that reconstruction procedure with tools of Computer Science, instead of trying to improve analysis of images that it outputs (which is what we do in the competition). What makes it impossible to apply more computer science to get really accurate images of kidneys? (as opposed to trying to recover vasculature from inaccurate images)</p>\n<p>One reason I can imagine is: imaging takes a time frame that is so long, that it is impossible to define the \"real accurate image\", since it changes during that time frame.</p>\n<p>So, I would appreciate if someone explained why obtaining more accurate images using CS applied to reconstruction problem is impossible, or less feasible than doing post-analysis. The reason I am concerned is that if it is possible, then all the fine data science work that is proposed for us to do could all be in vain, once accurate imaging becomes possible.</p>",
  "messages": [
    {
      "id": "2573361",
      "postDate": "12/24/2023 23:20:26",
      "content": "<p>As I understand, imaging methods basically throw something (like x-rays) at an object (like a kidney), then collect some signal resulting from that throwing (like received intensities of x-rays), from which they then try to reconstruct a 3-d image of the object.</p>\n<p>It seems natural to ask a question, why isn't it better to improve that reconstruction procedure with tools of Computer Science, instead of trying to improve analysis of images that it outputs (which is what we do in the competition). What makes it impossible to apply more computer science to get really accurate images of kidneys? (as opposed to trying to recover vasculature from inaccurate images)</p>\n<p>One reason I can imagine is: imaging takes a time frame that is so long, that it is impossible to define the \"real accurate image\", since it changes during that time frame.</p>\n<p>So, I would appreciate if someone explained why obtaining more accurate images using CS applied to reconstruction problem is impossible, or less feasible than doing post-analysis. The reason I am concerned is that if it is possible, then all the fine data science work that is proposed for us to do could all be in vain, once accurate imaging becomes possible.</p>",
      "rawMarkdown": "As I understand, imaging methods basically throw something (like x-rays) at an object (like a kidney), then collect some signal resulting from that throwing (like received intensities of x-rays), from which they then try to reconstruct a 3-d image of the object.\n\nIt seems natural to ask a question, why isn't it better to improve that reconstruction procedure with tools of Computer Science, instead of trying to improve analysis of images that it outputs (which is what we do in the competition). What makes it impossible to apply more computer science to get really accurate images of kidneys? (as opposed to trying to recover vasculature from inaccurate images)\n\nOne reason I can imagine is: imaging takes a time frame that is so long, that it is impossible to define the \"real accurate image\", since it changes during that time frame.\n\nSo, I would appreciate if someone explained why obtaining more accurate images using CS applied to reconstruction problem is impossible, or less feasible than doing post-analysis. The reason I am concerned is that if it is possible, then all the fine data science work that is proposed for us to do could all be in vain, once accurate imaging becomes possible.",
      "votes": null
    },
    {
      "id": "2592608",
      "postDate": "01/08/2024 17:15:09",
      "content": "<ol>\n<li>If it is about getting more out of a 'beam', I would argue if \"Hierarchical Phase-Contrast Tomography\" is not just doing that? So there is much more than 'throwing some x-ray'. I would assume all the process from beam production to final CT image actually uses all it can and optimized as the the level of technology is feasibly allowing.. So there is already a lot being done about reconstruction.</li>\n<li>Now detection (that something is there) and diagnosis/recognition (that it is our focus subject) are two different but of course not irrelevant ball parks.<br>\nIf you want to have a single machine do all these, it is fine, just embed your model and processor to this new machine and call it \"Hierarchical Phase-Contrast Tomography Super Machine\". <br>\nSo all is well. </li>\n</ol>",
      "rawMarkdown": "1. If it is about getting more out of a 'beam', I would argue if \"Hierarchical Phase-Contrast Tomography\" is not just doing that? So there is much more than 'throwing some x-ray'. I would assume all the process from beam production to final CT image actually uses all it can and optimized as the the level of technology is feasibly allowing.. So there is already a lot being done about reconstruction.\n2. Now detection (that something is there) and diagnosis/recognition (that it is our focus subject) are two different but of course not irrelevant ball parks.\n If you want to have a single machine do all these, it is fine, just embed your model and processor to this new machine and call it \"Hierarchical Phase-Contrast Tomography Super Machine\". \nSo all is well.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2592608,
      "author_name": "abdulkadirguner",
      "author_url": "",
      "post_date": "01/08/2024 17:15:09",
      "content": "<ol>\n<li>If it is about getting more out of a 'beam', I would argue if \"Hierarchical Phase-Contrast Tomography\" is not just doing that? So there is much more than 'throwing some x-ray'. I would assume all the process from beam production to final CT image actually uses all it can and optimized as the the level of technology is feasibly allowing.. So there is already a lot being done about reconstruction.</li>\n<li>Now detection (that something is there) and diagnosis/recognition (that it is our focus subject) are two different but of course not irrelevant ball parks.<br>\nIf you want to have a single machine do all these, it is fine, just embed your model and processor to this new machine and call it \"Hierarchical Phase-Contrast Tomography Super Machine\". <br>\nSo all is well. </li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2573361": "As I understand, imaging methods basically throw something (like x-rays) at an object (like a kidney), then collect some signal resulting from that throwing (like received intensities of x-rays), from which they then try to reconstruct a 3-d image of the object.\n\nIt seems natural to ask a question, why isn't it better to improve that reconstruction procedure with tools of Computer Science, instead of trying to improve analysis of images that it outputs (which is what we do in the competition). What makes it impossible to apply more computer science to get really accurate images of kidneys? (as opposed to trying to recover vasculature from inaccurate images)\n\nOne reason I can imagine is: imaging takes a time frame that is so long, that it is impossible to define the \"real accurate image\", since it changes during that time frame.\n\nSo, I would appreciate if someone explained why obtaining more accurate images using CS applied to reconstruction problem is impossible, or less feasible than doing post-analysis. The reason I am concerned is that if it is possible, then all the fine data science work that is proposed for us to do could all be in vain, once accurate imaging becomes possible.",
    "2592608": "1. If it is about getting more out of a 'beam', I would argue if \"Hierarchical Phase-Contrast Tomography\" is not just doing that? So there is much more than 'throwing some x-ray'. I would assume all the process from beam production to final CT image actually uses all it can and optimized as the the level of technology is feasibly allowing.. So there is already a lot being done about reconstruction.\n2. Now detection (that something is there) and diagnosis/recognition (that it is our focus subject) are two different but of course not irrelevant ball parks.\n If you want to have a single machine do all these, it is fine, just embed your model and processor to this new machine and call it \"Hierarchical Phase-Contrast Tomography Super Machine\". \nSo all is well."
  },
  "source": "meta"
}