{
  "id": 253297,
  "title": "How about this as a baseline?",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/253297",
  "author_name": "",
  "post_date": "2021-07-15T22:14:57.706651800Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Okay, so we have around 500 cases (or patients), and each case has 4 kinds of MRI: </p>\n<ul>\n<li>Fluid Attenuated Inversion Recovery (FLAIR)</li>\n<li>T1-weighted pre-contrast (T1w)</li>\n<li>T1-weighted post-contrast (T1Gd)</li>\n<li>T2-weighted (T2)</li>\n</ul>\n<p>And each MRI has a different length of Dicoms (varying per case). </p>\n<p>Our task? Classify patients to 0/1 binary which predicts MGMT Value.</p>\n<h2>Baseline IDEA</h2>\n<p>Convert case level levels to image-level labels such that each DICOM has a label. Train a simple image classifier.</p>\n<h2>Inference</h2>\n<p>Use this simple image classifier to make predictions per patient for every Dicom and take the mean? This becomes the prediction per patient. </p>\n<p>*Participating in this competition with the sole purpose of learning more about medical imaging and sharing ideas. Prizes are secondary ;)</p>",
  "messages": [
    {
      "id": "1389636",
      "postDate": "07/15/2021 22:14:57",
      "content": "<p>Okay, so we have around 500 cases (or patients), and each case has 4 kinds of MRI: </p>\n<ul>\n<li>Fluid Attenuated Inversion Recovery (FLAIR)</li>\n<li>T1-weighted pre-contrast (T1w)</li>\n<li>T1-weighted post-contrast (T1Gd)</li>\n<li>T2-weighted (T2)</li>\n</ul>\n<p>And each MRI has a different length of Dicoms (varying per case). </p>\n<p>Our task? Classify patients to 0/1 binary which predicts MGMT Value.</p>\n<h2>Baseline IDEA</h2>\n<p>Convert case level levels to image-level labels such that each DICOM has a label. Train a simple image classifier.</p>\n<h2>Inference</h2>\n<p>Use this simple image classifier to make predictions per patient for every Dicom and take the mean? This becomes the prediction per patient. </p>\n<p>*Participating in this competition with the sole purpose of learning more about medical imaging and sharing ideas. Prizes are secondary ;)</p>",
      "rawMarkdown": "Okay, so we have around 500 cases (or patients), and each case has 4 kinds of MRI: \n- Fluid Attenuated Inversion Recovery (FLAIR)\n- T1-weighted pre-contrast (T1w)\n- T1-weighted post-contrast (T1Gd)\n- T2-weighted (T2)\n\nAnd each MRI has a different length of Dicoms (varying per case). \n\nOur task? Classify patients to 0/1 binary which predicts MGMT Value.\n\n## Baseline IDEA\nConvert case level levels to image-level labels such that each DICOM has a label. Train a simple image classifier.\n\n## Inference\nUse this simple image classifier to make predictions per patient for every Dicom and take the mean? This becomes the prediction per patient. \n\n*Participating in this competition with the sole purpose of learning more about medical imaging and sharing ideas. Prizes are secondary ;)",
      "votes": null
    },
    {
      "id": "1389729",
      "postDate": "07/16/2021 02:50:24",
      "content": "<p>Great Idea <a href=\"https://www.kaggle.com/aroraaman\" target=\"_blank\">@aroraaman</a>, can you please help me with, how to load this data into a PyTorch data loader, or can you give me some resources to learn PyTorch custom data loader.</p>",
      "rawMarkdown": "Great Idea @aroraaman, can you please help me with, how to load this data into a PyTorch data loader, or can you give me some resources to learn PyTorch custom data loader.",
      "votes": null
    },
    {
      "id": "1392081",
      "postDate": "07/18/2021 11:03:17",
      "content": "<p>I also thought about labeling each image in the video with the video's label. Another interesting feature would be to add an LSTM on top of a CNN backbone so that information is transferred along the frames.<br>\nThis could be a solid baseline.</p>",
      "rawMarkdown": "I also thought about labeling each image in the video with the video's label. Another interesting feature would be to add an LSTM on top of a CNN backbone so that information is transferred along the frames.\nThis could be a solid baseline.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1389729,
      "author_name": "tharun2001",
      "author_url": "",
      "post_date": "07/16/2021 02:50:24",
      "content": "<p>Great Idea <a href=\"https://www.kaggle.com/aroraaman\" target=\"_blank\">@aroraaman</a>, can you please help me with, how to load this data into a PyTorch data loader, or can you give me some resources to learn PyTorch custom data loader.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1392081,
      "author_name": "alvarofbudria",
      "author_url": "",
      "post_date": "07/18/2021 11:03:17",
      "content": "<p>I also thought about labeling each image in the video with the video's label. Another interesting feature would be to add an LSTM on top of a CNN backbone so that information is transferred along the frames.<br>\nThis could be a solid baseline.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1389636": "Okay, so we have around 500 cases (or patients), and each case has 4 kinds of MRI: \n- Fluid Attenuated Inversion Recovery (FLAIR)\n- T1-weighted pre-contrast (T1w)\n- T1-weighted post-contrast (T1Gd)\n- T2-weighted (T2)\n\nAnd each MRI has a different length of Dicoms (varying per case). \n\nOur task? Classify patients to 0/1 binary which predicts MGMT Value.\n\n## Baseline IDEA\nConvert case level levels to image-level labels such that each DICOM has a label. Train a simple image classifier.\n\n## Inference\nUse this simple image classifier to make predictions per patient for every Dicom and take the mean? This becomes the prediction per patient. \n\n*Participating in this competition with the sole purpose of learning more about medical imaging and sharing ideas. Prizes are secondary ;)",
    "1389729": "Great Idea @aroraaman, can you please help me with, how to load this data into a PyTorch data loader, or can you give me some resources to learn PyTorch custom data loader.",
    "1392081": "I also thought about labeling each image in the video with the video's label. Another interesting feature would be to add an LSTM on top of a CNN backbone so that information is transferred along the frames.\nThis could be a solid baseline."
  },
  "source": "meta"
}