{
  "id": 451396,
  "title": "97th Place Solution for the RSNA 2023 Abdominal Trauma Detection",
  "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/451396",
  "author_name": "Yurnero",
  "post_date": "2023-10-28T14:48:42.232000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p><strong>Context</strong></p>\n<p>Business context: <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview</a><br>\nData context: <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data</a></p>\n<p><strong>Overview of the Approach</strong></p>\n<p>The final model is Resnet 2.5D + LSTM (1 layer with <code>hidden_size = 128</code>) was trained with <code>volume = (128, 64, 64)</code>, <code>window_step = 2</code> and <code>window_width = 3</code>.</p>\n<p><strong>Details of the submission</strong></p>\n<p>The <em>sigmoid</em> activation fuction was used for the output for <code>bowel_injury</code> and <code>extravasation_injury</code>, while <em>softmax</em> was used for the other targets</p>\n<p><strong>Sources</strong></p>\n<p><a href=\"https://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna\" target=\"_blank\">https://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna</a> — nice EDA notebook<br>\n<a href=\"https://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train\" target=\"_blank\">https://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train</a> — train EfficientNet on TPU<br>\n<a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449</a> — 1st place solution</p>",
  "messages": [
    {
      "id": 2502899,
      "postDate": "2023-10-28T14:48:42.233Z",
      "content": "<p><strong>Context</strong></p>\n<p>Business context: <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview</a><br>\nData context: <a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data</a></p>\n<p><strong>Overview of the Approach</strong></p>\n<p>The final model is Resnet 2.5D + LSTM (1 layer with <code>hidden_size = 128</code>) was trained with <code>volume = (128, 64, 64)</code>, <code>window_step = 2</code> and <code>window_width = 3</code>.</p>\n<p><strong>Details of the submission</strong></p>\n<p>The <em>sigmoid</em> activation fuction was used for the output for <code>bowel_injury</code> and <code>extravasation_injury</code>, while <em>softmax</em> was used for the other targets</p>\n<p><strong>Sources</strong></p>\n<p><a href=\"https://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna\" target=\"_blank\">https://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna</a> — nice EDA notebook<br>\n<a href=\"https://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train\" target=\"_blank\">https://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train</a> — train EfficientNet on TPU<br>\n<a href=\"https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449\" target=\"_blank\">https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449</a> — 1st place solution</p>",
      "rawMarkdown": "**Context**\n\nBusiness context: https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview\nData context: https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data\n\n\n**Overview of the Approach**\n\nThe final model is Resnet 2.5D + LSTM (1 layer with `hidden_size = 128`) was trained with `volume = (128, 64, 64)`, `window_step = 2` and `window_width = 3`.\n\n**Details of the submission**\n\nThe *sigmoid* activation fuction was used for the output for `bowel_injury` and `extravasation_injury`, while *softmax* was used for the other targets\n\n**Sources**\n\nhttps://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna — nice EDA notebook\nhttps://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train — train EfficientNet on TPU\nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449 — 1st place solution",
      "votes": 2
    },
    {
      "id": 2506518,
      "postDate": "2023-10-31T10:58:52.123Z",
      "content": "<p>Interesting</p>",
      "rawMarkdown": "Interesting"
    }
  ],
  "comments": [
    {
      "id": 2506518,
      "author_name": "Varun Balakrishna",
      "author_url": "",
      "post_date": "2023-10-31T10:58:52.123000",
      "content": "<p>Interesting</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2502899": "**Context**\n\nBusiness context: https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/overview\nData context: https://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/data\n\n\n**Overview of the Approach**\n\nThe final model is Resnet 2.5D + LSTM (1 layer with `hidden_size = 128`) was trained with `volume = (128, 64, 64)`, `window_step = 2` and `window_width = 3`.\n\n**Details of the submission**\n\nThe *sigmoid* activation fuction was used for the output for `bowel_injury` and `extravasation_injury`, while *softmax* was used for the other targets\n\n**Sources**\n\nhttps://www.kaggle.com/code/ayushs9020/understanding-the-competition-rsna — nice EDA notebook\nhttps://www.kaggle.com/code/awsaf49/rsna-atd-cnn-tpu-train — train EfficientNet on TPU\nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/447449 — 1st place solution",
    "2506518": "Interesting"
  }
}