{
  "id": 448282,
  "title": "Did anyone successfully utilize two aligned series?",
  "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/448282",
  "author_name": "yu4u",
  "post_date": "2023-10-19T00:27:46.587000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I believe that aligning two voxels from the same patient may contribute to performance improvement, especially in the recognition of extravasation.<br>\nDid anyone successfully utilize two aligned voxels?<br>\nI built a model with two aligned voxels as input but it did not work.<br>\n<a href=\"https://www.kaggle.com/code/ren4yu/rsna2023-align-two-voxels\" target=\"_blank\">Here is a sample notebook</a> that aligns two voxels for those challenging this direction.</p>\n<p>What I'm doing in this notebook is</p>\n<ul>\n<li>extract feature from each slice using <a href=\"https://github.com/facebookresearch/sscd-copy-detection\" target=\"_blank\">image copy detection model</a></li>\n<li>calculate similarity matrix between two voxels</li>\n<li>binarize the above matrix by taking max in horizontal or vertial direction</li>\n<li>detect line (= correspondence of two voxels) using Hough Transformation</li>\n</ul>\n<p>Some errors may be found in this notebook, but I have actually done crop as a preprocessing and it seemed to work well with all patient in that case.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb514a4da8bddf02955486934634deaa6%2F2023-10-19%209.25.56.png?generation=1697675196912324&amp;alt=media\" alt=\"\"></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb72e9a7c695bf35412e44ac86177537b%2F2023-10-19%209.26.13.png?generation=1697675213282799&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 2488012,
      "postDate": "2023-10-19T00:27:46.587Z",
      "content": "<p>I believe that aligning two voxels from the same patient may contribute to performance improvement, especially in the recognition of extravasation.<br>\nDid anyone successfully utilize two aligned voxels?<br>\nI built a model with two aligned voxels as input but it did not work.<br>\n<a href=\"https://www.kaggle.com/code/ren4yu/rsna2023-align-two-voxels\" target=\"_blank\">Here is a sample notebook</a> that aligns two voxels for those challenging this direction.</p>\n<p>What I'm doing in this notebook is</p>\n<ul>\n<li>extract feature from each slice using <a href=\"https://github.com/facebookresearch/sscd-copy-detection\" target=\"_blank\">image copy detection model</a></li>\n<li>calculate similarity matrix between two voxels</li>\n<li>binarize the above matrix by taking max in horizontal or vertial direction</li>\n<li>detect line (= correspondence of two voxels) using Hough Transformation</li>\n</ul>\n<p>Some errors may be found in this notebook, but I have actually done crop as a preprocessing and it seemed to work well with all patient in that case.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb514a4da8bddf02955486934634deaa6%2F2023-10-19%209.25.56.png?generation=1697675196912324&amp;alt=media\" alt=\"\"></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb72e9a7c695bf35412e44ac86177537b%2F2023-10-19%209.26.13.png?generation=1697675213282799&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I believe that aligning two voxels from the same patient may contribute to performance improvement, especially in the recognition of extravasation.\nDid anyone successfully utilize two aligned voxels?\nI built a model with two aligned voxels as input but it did not work.\n[Here is a sample notebook](https://www.kaggle.com/code/ren4yu/rsna2023-align-two-voxels) that aligns two voxels for those challenging this direction.\n\nWhat I'm doing in this notebook is\n\n- extract feature from each slice using [image copy detection model](https://github.com/facebookresearch/sscd-copy-detection)\n- calculate similarity matrix between two voxels\n- binarize the above matrix by taking max in horizontal or vertial direction\n- detect line (= correspondence of two voxels) using Hough Transformation\n\n\nSome errors may be found in this notebook, but I have actually done crop as a preprocessing and it seemed to work well with all patient in that case.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb514a4da8bddf02955486934634deaa6%2F2023-10-19%209.25.56.png?generation=1697675196912324&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb72e9a7c695bf35412e44ac86177537b%2F2023-10-19%209.26.13.png?generation=1697675213282799&alt=media)\n",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2488012": "I believe that aligning two voxels from the same patient may contribute to performance improvement, especially in the recognition of extravasation.\nDid anyone successfully utilize two aligned voxels?\nI built a model with two aligned voxels as input but it did not work.\n[Here is a sample notebook](https://www.kaggle.com/code/ren4yu/rsna2023-align-two-voxels) that aligns two voxels for those challenging this direction.\n\nWhat I'm doing in this notebook is\n\n- extract feature from each slice using [image copy detection model](https://github.com/facebookresearch/sscd-copy-detection)\n- calculate similarity matrix between two voxels\n- binarize the above matrix by taking max in horizontal or vertial direction\n- detect line (= correspondence of two voxels) using Hough Transformation\n\n\nSome errors may be found in this notebook, but I have actually done crop as a preprocessing and it seemed to work well with all patient in that case.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb514a4da8bddf02955486934634deaa6%2F2023-10-19%209.25.56.png?generation=1697675196912324&alt=media)\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F745525%2Fb72e9a7c695bf35412e44ac86177537b%2F2023-10-19%209.26.13.png?generation=1697675213282799&alt=media)\n"
  }
}