{
  "id": 13147,
  "title": "RGB or grayscale",
  "url": "/competitions/diabetic-retinopathy-detection/discussion/13147",
  "author_name": "",
  "post_date": "2015-03-30T18:34:02.740Z",
  "votes": 3,
  "comment_count": 3,
  "views": 1992,
  "content": "<p>The title kind of says it all. What are your opinions on this? Do you think it's possible to build a powerful model based on a grayscale representation of the data, or an RGB representation is required? In other words, could a color blind clinician do a good job at this? :)</p>",
  "messages": [
    {
      "id": "69057",
      "postDate": "03/30/2015 18:34:02",
      "content": "<p>The title kind of says it all. What are your opinions on this? Do you think it's possible to build a powerful model based on a grayscale representation of the data, or an RGB representation is required? In other words, could a color blind clinician do a good job at this? :)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "69076",
      "postDate": "03/30/2015 20:25:01",
      "content": "<p>Most of the publications I have read used the green channel for detecting hemorrhages, vessels and microaneurysms. Of course color information can be helpful too, but take into consideration that the appearing colors on the fundus images vary by&nbsp;racial groups.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "69136",
      "postDate": "03/31/2015 07:57:48",
      "content": "<p>The red channel is somewhat saturated, and the green channel&nbsp;contains more structural information. So it makes sense to use the green for segmentation / morphological operations, since most image processing tools work in greyscale only. However the colour itself may be important.</p>\n<p>I did not know that colour was related&nbsp;to ethnicity.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "69138",
      "postDate": "03/31/2015 08:15:50",
      "content": "<p>[Usher et al., 2003, Automated detection of diabetic retinopathy in digital retinal images: a tool for diabetic retinopathy screening]: <em>Intensity varied between images, and there were considerable colour variations due to differing ethnic origin.<br><br></em>Also there are some more publications mentioning this effect. Of course if we use the green channel and apply some pre-processing on it, this fact does not matter as much. :)</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 69076,
      "author_name": "zoltanfegyver",
      "author_url": "",
      "post_date": "03/30/2015 20:25:01",
      "content": "<p>Most of the publications I have read used the green channel for detecting hemorrhages, vessels and microaneurysms. Of course color information can be helpful too, but take into consideration that the appearing colors on the fundus images vary by&nbsp;racial groups.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 69136,
      "author_name": "hugues",
      "author_url": "",
      "post_date": "03/31/2015 07:57:48",
      "content": "<p>The red channel is somewhat saturated, and the green channel&nbsp;contains more structural information. So it makes sense to use the green for segmentation / morphological operations, since most image processing tools work in greyscale only. However the colour itself may be important.</p>\n<p>I did not know that colour was related&nbsp;to ethnicity.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 69138,
      "author_name": "zoltanfegyver",
      "author_url": "",
      "post_date": "03/31/2015 08:15:50",
      "content": "<p>[Usher et al., 2003, Automated detection of diabetic retinopathy in digital retinal images: a tool for diabetic retinopathy screening]: <em>Intensity varied between images, and there were considerable colour variations due to differing ethnic origin.<br><br></em>Also there are some more publications mentioning this effect. Of course if we use the green channel and apply some pre-processing on it, this fact does not matter as much. :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "69057": "",
    "69076": "",
    "69136": "",
    "69138": ""
  },
  "source": "meta"
}