{
  "id": 180699,
  "title": "Scan jpgs and Segmented lung jpgs dataset",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/180699",
  "author_name": "",
  "post_date": "2020-09-06T06:41:46.735693Z",
  "votes": 7,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Converted the DICOMs to jpg to make it easier to work with. I've jpg of raw scans and also lung segmentations generated by <a href=\"https://www.kaggle.com/aadhavvignesh/lung-segmentation-by-marker-controlled-watershed\" target=\"_blank\">marker-controlled watershed segmentation</a></p>\n<p>Dataset -  <a href=\"https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs\" target=\"_blank\">https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs</a> 4.2GB</p>\n<p>The reason I re-did this was to get the mapping from Hounsfield values to jpg pixel right. The datasets available now are normalizing the image using the <code>mean</code> and <code>std.dev</code> of each image, which will not provide a uniform mapping from Hounsfield values to jpg pixel values. FastAI deals with this problem by using frequency-based binning, which is what I've used.</p>\n<p>Kernel to generate Dataset - <a href=\"https://www.kaggle.com/gautham11/generating-scan-and-segemented-scan-jpg-fastai\" target=\"_blank\">Generating Scan and Segemented Scan jpg (fastAI)</a></p>\n<p>I'm working on adding support for GDCM, and including data cleaning steps. It will be really helpful if anyone can provide feedback on how to improve this.</p>",
  "messages": [
    {
      "id": "999947",
      "postDate": "09/06/2020 06:41:46",
      "content": "<p>Converted the DICOMs to jpg to make it easier to work with. I've jpg of raw scans and also lung segmentations generated by <a href=\"https://www.kaggle.com/aadhavvignesh/lung-segmentation-by-marker-controlled-watershed\" target=\"_blank\">marker-controlled watershed segmentation</a></p>\n<p>Dataset -  <a href=\"https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs\" target=\"_blank\">https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs</a> 4.2GB</p>\n<p>The reason I re-did this was to get the mapping from Hounsfield values to jpg pixel right. The datasets available now are normalizing the image using the <code>mean</code> and <code>std.dev</code> of each image, which will not provide a uniform mapping from Hounsfield values to jpg pixel values. FastAI deals with this problem by using frequency-based binning, which is what I've used.</p>\n<p>Kernel to generate Dataset - <a href=\"https://www.kaggle.com/gautham11/generating-scan-and-segemented-scan-jpg-fastai\" target=\"_blank\">Generating Scan and Segemented Scan jpg (fastAI)</a></p>\n<p>I'm working on adding support for GDCM, and including data cleaning steps. It will be really helpful if anyone can provide feedback on how to improve this.</p>",
      "rawMarkdown": "Converted the DICOMs to jpg to make it easier to work with. I've jpg of raw scans and also lung segmentations generated by [marker-controlled watershed segmentation](https://www.kaggle.com/aadhavvignesh/lung-segmentation-by-marker-controlled-watershed)\n\nDataset -  [https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs]( https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs) 4.2GB\n\nThe reason I re-did this was to get the mapping from Hounsfield values to jpg pixel right. The datasets available now are normalizing the image using the `mean` and `std.dev` of each image, which will not provide a uniform mapping from Hounsfield values to jpg pixel values. FastAI deals with this problem by using frequency-based binning, which is what I've used.\n\nKernel to generate Dataset - [Generating Scan and Segemented Scan jpg (fastAI)](https://www.kaggle.com/gautham11/generating-scan-and-segemented-scan-jpg-fastai)\n\nI'm working on adding support for GDCM, and including data cleaning steps. It will be really helpful if anyone can provide feedback on how to improve this.",
      "votes": null
    },
    {
      "id": "999954",
      "postDate": "09/06/2020 06:49:16",
      "content": "<p>thanks for sharing. i was doing it yesterday and it wasn't successful. </p>",
      "rawMarkdown": "thanks for sharing. i was doing it yesterday and it wasn't successful.",
      "votes": null
    },
    {
      "id": "1000048",
      "postDate": "09/06/2020 08:33:33",
      "content": "<p>keep up the great work <a href=\"https://www.kaggle.com/gautham11\" target=\"_blank\">@gautham11</a> </p>",
      "rawMarkdown": "keep up the great work @gautham11",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 999954,
      "author_name": "yimacs",
      "author_url": "",
      "post_date": "09/06/2020 06:49:16",
      "content": "<p>thanks for sharing. i was doing it yesterday and it wasn't successful. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1000048,
      "author_name": "maciejgronczynski",
      "author_url": "",
      "post_date": "09/06/2020 08:33:33",
      "content": "<p>keep up the great work <a href=\"https://www.kaggle.com/gautham11\" target=\"_blank\">@gautham11</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "999947": "Converted the DICOMs to jpg to make it easier to work with. I've jpg of raw scans and also lung segmentations generated by [marker-controlled watershed segmentation](https://www.kaggle.com/aadhavvignesh/lung-segmentation-by-marker-controlled-watershed)\n\nDataset -  [https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs]( https://www.kaggle.com/gautham11/pulmonaryfibrosisprogressionsegmentationjpgs) 4.2GB\n\nThe reason I re-did this was to get the mapping from Hounsfield values to jpg pixel right. The datasets available now are normalizing the image using the `mean` and `std.dev` of each image, which will not provide a uniform mapping from Hounsfield values to jpg pixel values. FastAI deals with this problem by using frequency-based binning, which is what I've used.\n\nKernel to generate Dataset - [Generating Scan and Segemented Scan jpg (fastAI)](https://www.kaggle.com/gautham11/generating-scan-and-segemented-scan-jpg-fastai)\n\nI'm working on adding support for GDCM, and including data cleaning steps. It will be really helpful if anyone can provide feedback on how to improve this.",
    "999954": "thanks for sharing. i was doing it yesterday and it wasn't successful.",
    "1000048": "keep up the great work @gautham11"
  },
  "source": "meta"
}