{
  "id": 264072,
  "title": "Are MRI types directly compareable?",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264072",
  "author_name": "",
  "post_date": "2021-08-11T02:33:19.063102300Z",
  "votes": 5,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I've never worked with MRI data and was wonder if different types FLAIR, T1w .. are directly comparable to each other.<br>\nWould really appreciate if someone can redirect me to any resources for reading about what these different scans </p>",
  "messages": [
    {
      "id": "1465335",
      "postDate": "08/11/2021 02:33:19",
      "content": "<p>I've never worked with MRI data and was wonder if different types FLAIR, T1w .. are directly comparable to each other.<br>\nWould really appreciate if someone can redirect me to any resources for reading about what these different scans </p>",
      "rawMarkdown": "I've never worked with MRI data and was wonder if different types FLAIR, T1w .. are directly comparable to each other.\nWould really appreciate if someone can redirect me to any resources for reading about what these different scans",
      "votes": null
    },
    {
      "id": "1466025",
      "postDate": "08/11/2021 09:23:52",
      "content": "<p>I summarise my naive and simplified understanding of the different modalities here: <a href=\"https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182\" target=\"_blank\">https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182</a><br>\nI am not an expert in the field at all though.</p>",
      "rawMarkdown": "I summarise my naive and simplified understanding of the different modalities here: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182\nI am not an expert in the field at all though.",
      "votes": null
    },
    {
      "id": "1469508",
      "postDate": "08/13/2021 01:44:05",
      "content": "<p>Dear Aryaman.&nbsp;</p>\n<p>I’m a neuroradiologist and can help with some practical intuitions for this challenge.<br>\nIn radiology we can acquire images using different physical parameters, called sequences. Usually each sequence is acquired separately.</p>\n<p>T1, T2 and FLAIR are the most basic and common sequences used for reading MRI. We have many others, but those 3 are the most basic and common ones.</p>\n<p>Starting with FLAIR, it is one of the most valuable sequences. Usually,&nbsp;expansive processes, just like tumors, promote edema in normal brain tissues. And FLAIR is a very good sequence to identify lesions and the edema around them.&nbsp;</p>\n<p>T2 is a similar sequence, but differently from FLAIR, the cerebrospinal fluid (CSF) will remain brighting. As edema is bright on both T2 and FLAIR sequences, the last one has the advantage of suppressing all CSF&nbsp;signals, leading to a better contrast between normal brain and lesion+edema. If we had to choose one sequence to evaluate lesion/edema, it would be FLAIR.&nbsp;</p>\n<p>For last, but not the least, we have T1. This sequence is good for evaluating the brain anatomy. In the T1 sequence, fat and blood (during some periods) appear bright. But T1 hides a very useful&nbsp;advantage:When you inject intravenous contrast (gadolinium), abnormal tissues will be bright. And lesions just like tumors and infections are avid for showing enhancement as they break the normal hematoencephalic barrier.&nbsp;</p>\n<p>Other sequences like SWI have shown better correlation with MGMT promoter mutation, but they are not as available as the basic T2, FLAIR and T1 pre and post gadolinium sequences.</p>\n<p>In fact, the MR physics and physiopathology behind all those biological processes are obviously much more complicated. But I hope this can bring some practical intuitions for developing your models.&nbsp;</p>\n<p>In this link, you can take a look at a glioblastoma, the same kind of tumor we are evaluating.<br>\n<a href=\"https://radiopaedia.org/cases/glioblastoma-nos-4\" target=\"_blank\">https://radiopaedia.org/cases/glioblastoma-nos-4</a></p>\n<p>Good luck</p>",
      "rawMarkdown": "Dear Aryaman. \n\nI’m a neuroradiologist and can help with some practical intuitions for this challenge.\nIn radiology we can acquire images using different physical parameters, called sequences. Usually each sequence is acquired separately.\n\nT1, T2 and FLAIR are the most basic and common sequences used for reading MRI. We have many others, but those 3 are the most basic and common ones.\n\nStarting with FLAIR, it is one of the most valuable sequences. Usually, expansive processes, just like tumors, promote edema in normal brain tissues. And FLAIR is a very good sequence to identify lesions and the edema around them. \n\nT2 is a similar sequence, but differently from FLAIR, the cerebrospinal fluid (CSF) will remain brighting. As edema is bright on both T2 and FLAIR sequences, the last one has the advantage of suppressing all CSF signals, leading to a better contrast between normal brain and lesion+edema. If we had to choose one sequence to evaluate lesion/edema, it would be FLAIR. \n\nFor last, but not the least, we have T1. This sequence is good for evaluating the brain anatomy. In the T1 sequence, fat and blood (during some periods) appear bright. But T1 hides a very useful advantage:When you inject intravenous contrast (gadolinium), abnormal tissues will be bright. And lesions just like tumors and infections are avid for showing enhancement as they break the normal hematoencephalic barrier. \n\nOther sequences like SWI have shown better correlation with MGMT promoter mutation, but they are not as available as the basic T2, FLAIR and T1 pre and post gadolinium sequences.\n\nIn fact, the MR physics and physiopathology behind all those biological processes are obviously much more complicated. But I hope this can bring some practical intuitions for developing your models. \n\nIn this link, you can take a look at a glioblastoma, the same kind of tumor we are evaluating.\nhttps://radiopaedia.org/cases/glioblastoma-nos-4\n\nGood luck",
      "votes": null
    },
    {
      "id": "1469788",
      "postDate": "08/13/2021 06:00:52",
      "content": "<p>Thank you for the practical intuition that you have provided. So if I am understood correctly, FLAIR  &gt; T1Gd &gt; T2 &gt; T1 in terms of detecting the tumor? </p>",
      "rawMarkdown": "Thank you for the practical intuition that you have provided. So if I am understood correctly, FLAIR  > T1Gd > T2 > T1 in terms of detecting the tumor?",
      "votes": null
    },
    {
      "id": "1470276",
      "postDate": "08/13/2021 12:25:52",
      "content": "<p>FLAIR is the most sensible sequence, but is not so specific as the edema surrounding the lesion will be bright just like the rest of the tumor.<br>\nT1Gd is more specific. Only tumoral tissue will enhance. Edema will not.</p>\n<p>But, some studies have shown that edema is relevant to distinct MGMT promoter methylation</p>\n<blockquote>\n  <p><em>MGMT promoter methylated glioblastoma is likely to show less edema</em><br>\n  <a href=\"http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711\" target=\"_blank\">http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711</a></p>\n</blockquote>\n<p>So, I would consider T1Gd to localize the tumor. And maybe the size of the edema, best seen at FLAIR, can help to distinguish MGMT 0/1.</p>\n<p>Cheers</p>",
      "rawMarkdown": "FLAIR is the most sensible sequence, but is not so specific as the edema surrounding the lesion will be bright just like the rest of the tumor.\nT1Gd is more specific. Only tumoral tissue will enhance. Edema will not.\n\nBut, some studies have shown that edema is relevant to distinct MGMT promoter methylation\n> *MGMT promoter methylated glioblastoma is likely to show less edema*\nhttp://www.ajnr.org/content/early/2018/07/12/ajnr.A5711\n\nSo, I would consider T1Gd to localize the tumor. And maybe the size of the edema, best seen at FLAIR, can help to distinguish MGMT 0/1.\n\nCheers",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1466025,
      "author_name": "smoschou55",
      "author_url": "",
      "post_date": "08/11/2021 09:23:52",
      "content": "<p>I summarise my naive and simplified understanding of the different modalities here: <a href=\"https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182\" target=\"_blank\">https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182</a><br>\nI am not an expert in the field at all though.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1469508,
      "author_name": "kuriki",
      "author_url": "",
      "post_date": "08/13/2021 01:44:05",
      "content": "<p>Dear Aryaman.&nbsp;</p>\n<p>I’m a neuroradiologist and can help with some practical intuitions for this challenge.<br>\nIn radiology we can acquire images using different physical parameters, called sequences. Usually each sequence is acquired separately.</p>\n<p>T1, T2 and FLAIR are the most basic and common sequences used for reading MRI. We have many others, but those 3 are the most basic and common ones.</p>\n<p>Starting with FLAIR, it is one of the most valuable sequences. Usually,&nbsp;expansive processes, just like tumors, promote edema in normal brain tissues. And FLAIR is a very good sequence to identify lesions and the edema around them.&nbsp;</p>\n<p>T2 is a similar sequence, but differently from FLAIR, the cerebrospinal fluid (CSF) will remain brighting. As edema is bright on both T2 and FLAIR sequences, the last one has the advantage of suppressing all CSF&nbsp;signals, leading to a better contrast between normal brain and lesion+edema. If we had to choose one sequence to evaluate lesion/edema, it would be FLAIR.&nbsp;</p>\n<p>For last, but not the least, we have T1. This sequence is good for evaluating the brain anatomy. In the T1 sequence, fat and blood (during some periods) appear bright. But T1 hides a very useful&nbsp;advantage:When you inject intravenous contrast (gadolinium), abnormal tissues will be bright. And lesions just like tumors and infections are avid for showing enhancement as they break the normal hematoencephalic barrier.&nbsp;</p>\n<p>Other sequences like SWI have shown better correlation with MGMT promoter mutation, but they are not as available as the basic T2, FLAIR and T1 pre and post gadolinium sequences.</p>\n<p>In fact, the MR physics and physiopathology behind all those biological processes are obviously much more complicated. But I hope this can bring some practical intuitions for developing your models.&nbsp;</p>\n<p>In this link, you can take a look at a glioblastoma, the same kind of tumor we are evaluating.<br>\n<a href=\"https://radiopaedia.org/cases/glioblastoma-nos-4\" target=\"_blank\">https://radiopaedia.org/cases/glioblastoma-nos-4</a></p>\n<p>Good luck</p>",
      "votes": null,
      "replies": [
        {
          "id": 1469788,
          "author_name": "ayuraj",
          "author_url": "",
          "post_date": "08/13/2021 06:00:52",
          "content": "<p>Thank you for the practical intuition that you have provided. So if I am understood correctly, FLAIR  &gt; T1Gd &gt; T2 &gt; T1 in terms of detecting the tumor? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1470276,
          "author_name": "kuriki",
          "author_url": "",
          "post_date": "08/13/2021 12:25:52",
          "content": "<p>FLAIR is the most sensible sequence, but is not so specific as the edema surrounding the lesion will be bright just like the rest of the tumor.<br>\nT1Gd is more specific. Only tumoral tissue will enhance. Edema will not.</p>\n<p>But, some studies have shown that edema is relevant to distinct MGMT promoter methylation</p>\n<blockquote>\n  <p><em>MGMT promoter methylated glioblastoma is likely to show less edema</em><br>\n  <a href=\"http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711\" target=\"_blank\">http://www.ajnr.org/content/early/2018/07/12/ajnr.A5711</a></p>\n</blockquote>\n<p>So, I would consider T1Gd to localize the tumor. And maybe the size of the edema, best seen at FLAIR, can help to distinguish MGMT 0/1.</p>\n<p>Cheers</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1465335": "I've never worked with MRI data and was wonder if different types FLAIR, T1w .. are directly comparable to each other.\nWould really appreciate if someone can redirect me to any resources for reading about what these different scans",
    "1466025": "I summarise my naive and simplified understanding of the different modalities here: https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/264182\nI am not an expert in the field at all though.",
    "1469508": "Dear Aryaman. \n\nI’m a neuroradiologist and can help with some practical intuitions for this challenge.\nIn radiology we can acquire images using different physical parameters, called sequences. Usually each sequence is acquired separately.\n\nT1, T2 and FLAIR are the most basic and common sequences used for reading MRI. We have many others, but those 3 are the most basic and common ones.\n\nStarting with FLAIR, it is one of the most valuable sequences. Usually, expansive processes, just like tumors, promote edema in normal brain tissues. And FLAIR is a very good sequence to identify lesions and the edema around them. \n\nT2 is a similar sequence, but differently from FLAIR, the cerebrospinal fluid (CSF) will remain brighting. As edema is bright on both T2 and FLAIR sequences, the last one has the advantage of suppressing all CSF signals, leading to a better contrast between normal brain and lesion+edema. If we had to choose one sequence to evaluate lesion/edema, it would be FLAIR. \n\nFor last, but not the least, we have T1. This sequence is good for evaluating the brain anatomy. In the T1 sequence, fat and blood (during some periods) appear bright. But T1 hides a very useful advantage:When you inject intravenous contrast (gadolinium), abnormal tissues will be bright. And lesions just like tumors and infections are avid for showing enhancement as they break the normal hematoencephalic barrier. \n\nOther sequences like SWI have shown better correlation with MGMT promoter mutation, but they are not as available as the basic T2, FLAIR and T1 pre and post gadolinium sequences.\n\nIn fact, the MR physics and physiopathology behind all those biological processes are obviously much more complicated. But I hope this can bring some practical intuitions for developing your models. \n\nIn this link, you can take a look at a glioblastoma, the same kind of tumor we are evaluating.\nhttps://radiopaedia.org/cases/glioblastoma-nos-4\n\nGood luck",
    "1469788": "Thank you for the practical intuition that you have provided. So if I am understood correctly, FLAIR  > T1Gd > T2 > T1 in terms of detecting the tumor?",
    "1470276": "FLAIR is the most sensible sequence, but is not so specific as the edema surrounding the lesion will be bright just like the rest of the tumor.\nT1Gd is more specific. Only tumoral tissue will enhance. Edema will not.\n\nBut, some studies have shown that edema is relevant to distinct MGMT promoter methylation\n> *MGMT promoter methylated glioblastoma is likely to show less edema*\nhttp://www.ajnr.org/content/early/2018/07/12/ajnr.A5711\n\nSo, I would consider T1Gd to localize the tumor. And maybe the size of the edema, best seen at FLAIR, can help to distinguish MGMT 0/1.\n\nCheers"
  },
  "source": "meta"
}