{
  "fetched_at": "2026-10-24T22:35:50Z",
  "source_url": "https://www.kaggle.com/competitions/covid-segmentation/data",
  "pages": [
    {
      "name": "rules",
      "content": "It is not a real competition. But it is an important one! So:\n\n - Don't cheat!\n - Apply yourself!\n - Have fun!\n - And share your code!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "data-description",
      "content": "The data was kindly provided by [medicalsegmentation.com](http://medicalsegmentation.com/covid19/). They are two radiologists from Oslo, who've done plenty of work scraping and segmenting CT images. Also, they have made a [great product](https://www.medseg.ai/), where you can segment CT images in a browser. The magic of their tool is that it has several pretrained Unets that run in offline mode (no user data is going out) and make your work a hundred times faster.\nAn additional contribution of lung masks was made by Johannes Hofmanninger. His approach on [Github](https://github.com/JoHof/lungmask).\n\n## Dataset consists of two parts:\n\n### Medseg part\nThis is a dataset of 100 axial CT images from >40 patients with COVID-19 that were converted from openly accessible JPG images found [HERE](https://www.sirm.org/en/). The conversion process is described in detail in the following blogpost: [Covid-19 radiology — data collection and preparation for Artificial Intelligence](https://medium.com/@hbjenssen/covid-19-radiology-data-collection-and-preparation-for-artificial-intelligence-4ecece97bb5b)\n\n**images_medseg.npy**  - training images – 100 slices 512x512 size<br>\n**masks_medseg.npy** - training masks – 100 masks with 4 channels: (0 - \"ground glass\", 1 - \"consolidations\", 2 - \"lungs other\", 3 - \"background\" )<br>\n**test_images_medseg.npy** - test images – 10 slices  512x512 size\n\n### Radiopedia part\nSegmented 9 axial volumetric CTs from [Radiopaedia](https://radiopaedia.org/articles/covid-19-3).  This dataset includes whole volumes and includes, therefore, both positive and negative slices (373 out of the total of 829 slices have been evaluated by a radiologist as positive and segmented). These volumes are converted and normalized in a similar way as above.\n\n**images_radiopedia.npy** - training images – 829 slices 512x512 size<br>\n**masks_radiopedia.npy** - training masks – 829 masks with 4 channels: (0 - \"ground glass\", 1 - \"consolidations\", 2 - \"lungs other\", 3 - \"background\" )\n\n## Predictions\nParticipants should predict 10 masks corresponding to test_images_medseg.npy with two classes: 0 - \"ground glass\", 1 - \"consolidations\".\n\n## Use of other datasets:\nParticipants can use any data available if they share it with others via posting a link in Discussion<br>\nFor example, a good source of labeled COVID-19 CT data is available here: [https://mosmed.ai](https://mosmed.ai)",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "For COVID-19 patients the common stage in diagnosis is computer tomography (CT). A radiologist is often asked to estimate the extent of damage with respect to lung volume. It is a time-consuming procedure, because radiologist should look through all axial slices on CT and segment each of them.\n\nThere are some COVID-specific findings: [https://radiologyassistant.nl/chest/covid-19-corads-classification](https://radiologyassistant.nl/chest/covid-19-corads-classification). In this challenge you are asked to segment \"ground-glass\" and \"consolidation\".\n\nThe goal of this challenge is to help companies, who are building CT-specific software, come up with better solutions.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "## Metric\nSubmissions are evaluated on pixel-wise [F1-score](https://en.wikipedia.org/wiki/F1_score). F1-score is averaged across all pixels of 10 test images and two classes. Similarly, as it is a classification challenge\n\n## File Format\nYour submission file should be in `csv` format, with a header and columns names: `Id,Predicted`.\n\nIf you have numpy array `test_masks_prediction` with shape `(10, 512, 512, 2)` - it can be converted to submission the following way:\n\n```\nimport pandas as pd\n\npd.DataFrame(\n             data=np.stack((np.arange(len(test_masks_prediction.ravel())), \n                                        test_masks_prediction.ravel().astype(int)),\n                                       axis=-1), \n             columns=['Id', 'Expected'])\\\n.set_index('Id').to_csv('submission.csv')\n```",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}