{
  "fetched_at": "2026-10-19T06:35:13Z",
  "source_url": "https://www.kaggle.com/competitions/dlai3-phase3/data",
  "pages": [
    {
      "name": "data-description",
      "content": "<p>This challenge has two multi-class datasets: one labelled and one unlabelled.</p>\n<h2>File descriptions</h2>\n<ul>\n<li><strong>DLAI3_Phase3.zip</strong> - the multi-class training set containing three folders of no-finding, thorax-diseases, and covid</li>\n<li><strong>sampleSubmission.csv</strong> - a sample submission file in the correct format</li>\n<li><strong>VALIDATE.zip</strong> - unlabelled set to be tested and submit in form of the sampleSubmission.csv which is two columns with first column as the filename and the second being \"0\" for No-Finding; \"1\" for Thorax-Diseases; and \"2\" for COVID-19.</li>\n</ul>\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "This is the Phase 3 challenge from the Hackathon of the 3rd Deep Learning and AI Summer/Winter School (DLAI3). More details about the previous challenge can be found at https://www.kaggle.com/c/dlai3/.\n\nMedical image analysis has continually been an area of prominent and growing importance, in both the research and application aspects. However, it is important to realize that the algorithms, methodology, as well as the source of data, need to be strictly scrutinized. Phase 3 is a challenge that provides a mixed dataset of publicly available COVID-19 chest x-ray images from various sources, a pneumonia dataset obtained from a study on children, partially verified thorax CXR images of adults from NIH, and unverified no-finding images from NIH. See below for the sources of these data.\n\n<h2>Goal</h2>\n<p>The goal is to provide the participants a chance to test their skills to come up with a good performance model in turns of the macro F1 score to classify COVID-19 chest x-ray images. The Phase 3 DLAI3 Hackathon will award the top three entries that have a full paper submission to the 11th International Conference on Computational Systems-Biology and Bioinformatics (<a href=\"http://www.csbio.org/2020/\">CSBio2020</a>), under the theme of CHAT-2020 (COVID-19 Health, Analytics, and Technologies). The top three individuals/teams will be provided with complimentary registration for their full paper submission to the ACM ICPS Proceedings of CSBio2020 and will be invited to submit an extended work to a post-conference journal special issue.</p>\n\n<h2>Dataset citation</h2>\nPlease cite the kaggle dataset as follows: Jonathan H. Chan, “DLAI3 Hackathon Phase3 COVID-19 CXR Challenge.” Kaggle, doi: [10.34740/KAGGLE/DSV/1347344](https://doi.org/10.34740/KAGGLE/DSV/1347344).\n\n<h1>Acknowledgments</h1>\n\nThe following are the sources and related publications for this dataset:\n\n1) Pneumonia CXR images of children retrieved from Mendeley at:\n[https://data.mendeley.com/datasets/rscbjbr9sj/2\n](https://data.mendeley.com/datasets/rscbjbr9sj/2)\nRelated publication:\nhttps://www.cell.com/cell/fulltext/S0092-8674(18)30154-5\n\n2) Cropped COVID-19 + selected images from the following (retrieved July 17, 2020):\n[https://github.com/agchung/Actualmed-COVID-chestxray-dataset\n](https://github.com/agchung/Actualmed-COVID-chestxray-dataset)\nRelated publication:\n[https://arxiv.org/abs/2003.09871\n](https://arxiv.org/abs/2003.09871)v4 Mon, 11 May 2020\n\n3) COVID-19 chest x-ray images from:\n[https://github.com/ieee8023/covid-chestxray-dataset\n](https://github.com/ieee8023/covid-chestxray-dataset)(Retrieved July 13, 2020)\n\nImages:\n[https://github.com/ieee8023/covid-chestxray-dataset/blob/master/images\n](https://github.com/ieee8023/covid-chestxray-dataset/blob/master/images)\nMetadata:\n[https://github.com/ieee8023/covid-chestxray-dataset/blob/master/metadata.csv\n](https://github.com/ieee8023/covid-chestxray-dataset/blob/master/metadata.csv)\n4) Selected images from:\n[https://nihcc.app.box.com/v/ChestXray-NIHCC\n](https://nihcc.app.box.com/v/ChestXray-NIHCC)\nPublication:\nChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on\nWeakly-Supervised Classification and Localization of Common Thorax Diseases\n\n[https://arxiv.org/abs/1705.02315\n](https://arxiv.org/abs/1705.02315)[v5] Thu, 14 Dec 2017\n\nSemi-verified by:\n[https://lukeoakdenrayner.wordpress.com/2017/12/18/the-chestxray14-dataset-problems/comment-page-1/](https://lukeoakdenrayner.wordpress.com/2017/12/18/the-chestxray14-dataset-problems/comment-page-1/)",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "<p>The evaluation metric is the F1 score, which combines the statistical measures of precision p and recall r. Precision is the ratio of true positives (tp) to all predicted positives (tp + fp). Recall is the ratio of true positives to all actual positives (tp + fn). The F1 score is given by:</p>\n<p>\\[ F1 = 2\\frac{p \\cdot r}{p+r}\\ \\ \\mathrm{where}\\ \\ p = \\frac{tp}{tp+fp},\\ \\ r = \\frac{tp}{tp+fn} \\]</p>\n<p>The F1 metric weights recall and precision equally, and a good retrieval algorithm will maximize both precision and recall simultaneously. Thus, moderately good performance on both will be favored over extremely good performance on one and poor performance on the other. For multi-class, the evaluation score is based on the macro average of all classes</p>\n<h2>Submission Format</h2>\n<p><strong>For every author in the dataset</strong>, submission files should contain two columns: unlabelled image name and predicted label as \"0\" for No-Finding; \"1\" for Thorax-Disease (other than COVID-19); and \"2\" for COVID-19. </p>\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "By participating in this challenge, the participants acknowledges that the datasets are provided for the purpose of the competition only. There is no claim to ownership/authenticity/validity of the datasets and annotations. Please do not rely on the datasets provided here for diagnosis purpose without detailed validation. Finally, the participants are encouraged to submit a quality publication, preferrably to CSBio2020 (http://www.csbio.org/2020/), based on their results from this challenge. Any publications must cite the following: Jonathan H. Chan, “DLAI3 Hackathon Phase3 COVID-19 CXR Challenge.” Kaggle, doi: [10.34740/KAGGLE/DSV/1347344](https://doi.org/10.34740/KAGGLE/DSV/1347344).\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}