{
  "id": 371081,
  "title": "what are some good DICOM datasets we can use for pretraining?",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/371081",
  "author_name": "Radek Osmulski",
  "post_date": "2022-12-07T21:19:37.932000",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey!</p>\n<p>The low incidence of the positive class makes training on this dataset really hard.</p>\n<p>One way to address this could be training on a rich DICOM dataset to teach our model how to read those mammography representations.</p>\n<p>What is a good dataset that in your mind could be used for pretraining? It doesn't have to be a mammography dataset (would be a plus though) but just a good dataset containing DICOM images would work as well.</p>\n<p>Any ideas would be greatly appreciated 🙏 Thank you!</p>",
  "messages": [
    {
      "id": 2058368,
      "postDate": "2022-12-07T21:19:37.933Z",
      "content": "<p>Hey!</p>\n<p>The low incidence of the positive class makes training on this dataset really hard.</p>\n<p>One way to address this could be training on a rich DICOM dataset to teach our model how to read those mammography representations.</p>\n<p>What is a good dataset that in your mind could be used for pretraining? It doesn't have to be a mammography dataset (would be a plus though) but just a good dataset containing DICOM images would work as well.</p>\n<p>Any ideas would be greatly appreciated 🙏 Thank you!</p>",
      "rawMarkdown": "Hey!\n\nThe low incidence of the positive class makes training on this dataset really hard.\n\nOne way to address this could be training on a rich DICOM dataset to teach our model how to read those mammography representations.\n\nWhat is a good dataset that in your mind could be used for pretraining? It doesn't have to be a mammography dataset (would be a plus though) but just a good dataset containing DICOM images would work as well.\n\nAny ideas would be greatly appreciated 🙏 Thank you!",
      "votes": 4
    },
    {
      "id": 2058382,
      "postDate": "2022-12-07T21:39:11.740Z",
      "content": "<p>DDSM has been uploaded to kaggle here - </p>\n<p><a href=\"https://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset\" target=\"_blank\">https://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset</a></p>\n<p>It has csv files which outline tumors as well, so that might be useful perhaps.</p>\n<p>Learn more about it here - <a href=\"https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629\" target=\"_blank\">https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629</a></p>",
      "rawMarkdown": "DDSM has been uploaded to kaggle here - \n\nhttps://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset\n\nIt has csv files which outline tumors as well, so that might be useful perhaps.\n\nLearn more about it here - https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629\n",
      "votes": 2,
      "replies": [
        {
          "id": 2058391,
          "postDate": "2022-12-07T22:05:57Z",
          "content": "<p>Thank you <a href=\"https://www.kaggle.com/kaggleqrdl\" target=\"_blank\">@kaggleqrdl</a>! 🙏 That is great to know! </p>\n<p>Seems to be of a decent size - 2600 patients!</p>\n<p>BTW whether the dataset is on kaggle or not is of secondary importance to me, can be anywhere 🙂</p>\n<p>But thanks for sharing, this is great! The extra labels can also be utilized as well in some way most likely!</p>",
          "rawMarkdown": "Thank you @kaggleqrdl! 🙏 That is great to know! \n\nSeems to be of a decent size - 2600 patients!\n\nBTW whether the dataset is on kaggle or not is of secondary importance to me, can be anywhere 🙂\n\nBut thanks for sharing, this is great! The extra labels can also be utilized as well in some way most likely!",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2058382,
      "author_name": "@kaggleqrdl",
      "author_url": "",
      "post_date": "2022-12-07T21:39:11.740000",
      "content": "<p>DDSM has been uploaded to kaggle here - </p>\n<p><a href=\"https://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset\" target=\"_blank\">https://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset</a></p>\n<p>It has csv files which outline tumors as well, so that might be useful perhaps.</p>\n<p>Learn more about it here - <a href=\"https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629\" target=\"_blank\">https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629</a></p>",
      "votes": 2,
      "replies": [
        {
          "id": 2058391,
          "author_name": "Radek Osmulski",
          "author_url": "",
          "post_date": "2022-12-07T22:05:57",
          "content": "<p>Thank you <a href=\"https://www.kaggle.com/kaggleqrdl\" target=\"_blank\">@kaggleqrdl</a>! 🙏 That is great to know! </p>\n<p>Seems to be of a decent size - 2600 patients!</p>\n<p>BTW whether the dataset is on kaggle or not is of secondary importance to me, can be anywhere 🙂</p>\n<p>But thanks for sharing, this is great! The extra labels can also be utilized as well in some way most likely!</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2058368": "Hey!\n\nThe low incidence of the positive class makes training on this dataset really hard.\n\nOne way to address this could be training on a rich DICOM dataset to teach our model how to read those mammography representations.\n\nWhat is a good dataset that in your mind could be used for pretraining? It doesn't have to be a mammography dataset (would be a plus though) but just a good dataset containing DICOM images would work as well.\n\nAny ideas would be greatly appreciated 🙏 Thank you!",
    "2058382": "DDSM has been uploaded to kaggle here - \n\nhttps://www.kaggle.com/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset\n\nIt has csv files which outline tumors as well, so that might be useful perhaps.\n\nLearn more about it here - https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=22516629\n"
  }
}