{
  "id": 446033,
  "title": "Criteria for healthy/unhealthy patches ?",
  "url": "/competitions/UBC-OCEAN/discussion/446033",
  "author_name": "Ziri",
  "post_date": "2023-10-10T03:01:44.682000",
  "votes": 4,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Let's say I divided the image into patches, is it possible to know the healthy/unhealthy patches using some kind of rules?</p>",
  "messages": [
    {
      "id": 2477547,
      "postDate": "2023-10-11T11:30:04.160Z",
      "content": "<p>I'm also experimenting with creating patches and filtering them by their content. I remove any patches that are mostly white (contains no cellular material), mostly black (removed areas), and mostly red (blood). Then I try to include patches with more pixels that are blue to purple because that is the stained color of DNA, and cancer cells generally have a lot of DNA.</p>",
      "rawMarkdown": "I'm also experimenting with creating patches and filtering them by their content. I remove any patches that are mostly white (contains no cellular material), mostly black (removed areas), and mostly red (blood). Then I try to include patches with more pixels that are blue to purple because that is the stained color of DNA, and cancer cells generally have a lot of DNA.",
      "votes": 12,
      "replies": [
        {
          "id": 2482593,
          "postDate": "2023-10-15T06:50:26.413Z",
          "content": "<ul>\n<li>Divide HxW image into n hxw patches</li>\n<li>Rank patches by \"information\"</li>\n<li>Use top k patches</li>\n</ul>\n<p>I think this is a very common approach in this field and it works pretty decently.</p>",
          "rawMarkdown": "* Divide HxW image into n hxw patches\n* Rank patches by \"information\"\n* Use top k patches\n\nI think this is a very common approach in this field and it works pretty decently.",
          "votes": 5
        },
        {
          "id": 2499735,
          "postDate": "2023-10-26T07:50:00.543Z",
          "content": "<p>That is a nice observation and could help for my training from tiles when I would not take only random but more likely DNA-rich samples. Just the preprocessing could take some time… <a href=\"https://kaggle.com/code/jirkaborovec/cancer-subtype-tiles-w-lightning-timm-models\" target=\"_blank\">https://kaggle.com/code/jirkaborovec/cancer-subtype-tiles-w-lightning-timm-models</a></p>",
          "rawMarkdown": "That is a nice observation and could help for my training from tiles when I would not take only random but more likely DNA-rich samples. Just the preprocessing could take some time... https://kaggle.com/code/jirkaborovec/cancer-subtype-tiles-w-lightning-timm-models",
          "votes": 2
        },
        {
          "id": 2512559,
          "postDate": "2023-11-04T16:36:32.323Z",
          "content": "<p>Do you have any source code for tiles selections?</p>",
          "rawMarkdown": "Do you have any source code for tiles selections?\n"
        }
      ]
    },
    {
      "id": 2475585,
      "postDate": "2023-10-10T03:01:44.683Z",
      "content": "<p>Let's say I divided the image into patches, is it possible to know the healthy/unhealthy patches using some kind of rules?</p>",
      "rawMarkdown": "Let's say I divided the image into patches, is it possible to know the healthy/unhealthy patches using some kind of rules?\n\n",
      "votes": 4
    }
  ],
  "comments": [
    {
      "id": 2477547,
      "author_name": "Noli Alonso",
      "author_url": "",
      "post_date": "2023-10-11T11:30:04.160000",
      "content": "<p>I'm also experimenting with creating patches and filtering them by their content. I remove any patches that are mostly white (contains no cellular material), mostly black (removed areas), and mostly red (blood). Then I try to include patches with more pixels that are blue to purple because that is the stained color of DNA, and cancer cells generally have a lot of DNA.</p>",
      "votes": 12,
      "replies": [
        {
          "id": 2482593,
          "author_name": "Gunes Evitan",
          "author_url": "",
          "post_date": "2023-10-15T06:50:26.413000",
          "content": "<ul>\n<li>Divide HxW image into n hxw patches</li>\n<li>Rank patches by \"information\"</li>\n<li>Use top k patches</li>\n</ul>\n<p>I think this is a very common approach in this field and it works pretty decently.</p>",
          "votes": 5,
          "replies": []
        },
        {
          "id": 2499735,
          "author_name": "Jirka",
          "author_url": "",
          "post_date": "2023-10-26T07:50:00.543000",
          "content": "<p>That is a nice observation and could help for my training from tiles when I would not take only random but more likely DNA-rich samples. Just the preprocessing could take some time… <a href=\"https://kaggle.com/code/jirkaborovec/cancer-subtype-tiles-w-lightning-timm-models\" target=\"_blank\">https://kaggle.com/code/jirkaborovec/cancer-subtype-tiles-w-lightning-timm-models</a></p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 2512559,
          "author_name": "DungNv1714",
          "author_url": "",
          "post_date": "2023-11-04T16:36:32.323000",
          "content": "<p>Do you have any source code for tiles selections?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2477547": "I'm also experimenting with creating patches and filtering them by their content. I remove any patches that are mostly white (contains no cellular material), mostly black (removed areas), and mostly red (blood). Then I try to include patches with more pixels that are blue to purple because that is the stained color of DNA, and cancer cells generally have a lot of DNA.",
    "2475585": "Let's say I divided the image into patches, is it possible to know the healthy/unhealthy patches using some kind of rules?\n\n"
  }
}