{
  "id": 159143,
  "title": "Different malignant images in ISIC 2019",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/159143",
  "author_name": "",
  "post_date": "2020-06-16T14:58:36.515167900Z",
  "votes": 7,
  "comment_count": 8,
  "views": 0,
  "content": "<p>External Datasets contain malignant images, which are different from melanomas: basal cell carcinoma and spinal cell carcinoma. When we use them, should the 'target' of the images only be '1' for the case of melanomas?\nIn all external dataset used here (from Shonenkow, Chris Deotte), the targets were set to '1' , when the image shows melanomas, but not in the case of basal cell carcinomas. \nThe ISIC 2019 dataset contains more than 8000 basal cell carcinomas, which are definitely malignant! \nThe question is, if the original dataset (and the test set) contains only melanomas as malignant images (because this is a Melanoma Classification Competition)!</p>",
  "messages": [
    {
      "id": "888774",
      "postDate": "06/16/2020 14:58:36",
      "content": "<p>External Datasets contain malignant images, which are different from melanomas: basal cell carcinoma and spinal cell carcinoma. When we use them, should the 'target' of the images only be '1' for the case of melanomas?\nIn all external dataset used here (from Shonenkow, Chris Deotte), the targets were set to '1' , when the image shows melanomas, but not in the case of basal cell carcinomas. \nThe ISIC 2019 dataset contains more than 8000 basal cell carcinomas, which are definitely malignant! \nThe question is, if the original dataset (and the test set) contains only melanomas as malignant images (because this is a Melanoma Classification Competition)!</p>",
      "rawMarkdown": "External Datasets contain malignant images, which are different from melanomas: basal cell carcinoma and spinal cell carcinoma. When we use them, should the 'target' of the images only be '1' for the case of melanomas?\nIn all external dataset used here (from Shonenkow, Chris Deotte), the targets were set to '1' , when the image shows melanomas, but not in the case of basal cell carcinomas. \nThe ISIC 2019 dataset contains more than 8000 basal cell carcinomas, which are definitely malignant! \nThe question is, if the original dataset (and the test set) contains only melanomas as malignant images (because this is a Melanoma Classification Competition)!",
      "votes": null
    },
    {
      "id": "890483",
      "postDate": "06/17/2020 14:25:14",
      "content": "<blockquote>\n  <p>basal cell carcinomas, which are definitely malignant</p>\n</blockquote>\n\n<p>This is completely false. See the article. <a href=\"https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html\">https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html</a></p>",
      "rawMarkdown": "&gt;  basal cell carcinomas, which are definitely malignant\n\nThis is completely false. See the article. https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html",
      "votes": null
    },
    {
      "id": "890575",
      "postDate": "06/17/2020 15:21:29",
      "content": "<p>As I understand these cell are malignant but not melanoma. When I used them for training I set the target to 0</p>",
      "rawMarkdown": "As I understand these cell are malignant but not melanoma. When I used them for training I set the target to 0",
      "votes": null
    },
    {
      "id": "890597",
      "postDate": "06/17/2020 15:40:20",
      "content": "<p>carcinoma = a cancer arising in the epithelial tissue of the skin or of the lining of the internal organs.</p>\n\n<p>cancer = malignant tumor</p>\n\n<p>So that statement is completely true. Nevertheless, this competition is about detecting melanomas as far as I know.</p>",
      "rawMarkdown": "carcinoma = a cancer arising in the epithelial tissue of the skin or of the lining of the internal organs.\n\ncancer = malignant tumor\n\nSo that statement is completely true. Nevertheless, this competition is about detecting melanomas as far as I know.",
      "votes": null
    },
    {
      "id": "890756",
      "postDate": "06/17/2020 17:18:44",
      "content": "<p>In the ISIC 2019 dataset are three different malignant skin lesions: Melanoma (MEL), basal cell carcinoma (BCC) and spinal cell carcinoma (SCC). \nAbout this, I am absolute sure. I work as medical doctor, deep learning is only a hobby for me.\nI think, Gilles Vandewiele is correct, that we only have to predict melanomas. But when not, our models forget to learn more than 8000 BCC an 800 SCC! This would make a big difference!</p>",
      "rawMarkdown": "In the ISIC 2019 dataset are three different malignant skin lesions: Melanoma (MEL), basal cell carcinoma (BCC) and spinal cell carcinoma (SCC). \nAbout this, I am absolute sure. I work as medical doctor, deep learning is only a hobby for me.\nI think, Gilles Vandewiele is correct, that we only have to predict melanomas. But when not, our models forget to learn more than 8000 BCC an 800 SCC! This would make a big difference!",
      "votes": null
    },
    {
      "id": "890927",
      "postDate": "06/17/2020 19:10:55",
      "content": "<p>Is the symptomatology for BCC and SCC more similar to MEL than to benign lesions such as nevi? If that is the case, then adding those as negative examples could confuse the model indeed (especially if these are not present within the public/private test set)!</p>",
      "rawMarkdown": "Is the symptomatology for BCC and SCC more similar to MEL than to benign lesions such as nevi? If that is the case, then adding those as negative examples could confuse the model indeed (especially if these are not present within the public/private test set)!",
      "votes": null
    },
    {
      "id": "891901",
      "postDate": "06/18/2020 14:19:59",
      "content": "<p>Does ISIC 2020 contain basal cell carcinomas? </p>",
      "rawMarkdown": "Does ISIC 2020 contain basal cell carcinomas?",
      "votes": null
    },
    {
      "id": "891990",
      "postDate": "06/18/2020 15:24:22",
      "content": "<p>Basal Cell Carcinomas and Squamous Cell Carcinomas are not considered Melanoma and would not be considered target = 1 for the purposes of this challenge. </p>",
      "rawMarkdown": "Basal Cell Carcinomas and Squamous Cell Carcinomas are not considered Melanoma and would not be considered target = 1 for the purposes of this challenge.",
      "votes": null
    },
    {
      "id": "892290",
      "postDate": "06/18/2020 19:18:46",
      "content": "<p>Oh thank you very much for this clarification!</p>",
      "rawMarkdown": "Oh thank you very much for this clarification!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 890483,
      "author_name": "nroman",
      "author_url": "",
      "post_date": "06/17/2020 14:25:14",
      "content": "<blockquote>\n  <p>basal cell carcinomas, which are definitely malignant</p>\n</blockquote>\n\n<p>This is completely false. See the article. <a href=\"https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html\">https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 890597,
          "author_name": "group16",
          "author_url": "",
          "post_date": "06/17/2020 15:40:20",
          "content": "<p>carcinoma = a cancer arising in the epithelial tissue of the skin or of the lining of the internal organs.</p>\n\n<p>cancer = malignant tumor</p>\n\n<p>So that statement is completely true. Nevertheless, this competition is about detecting melanomas as far as I know.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 890575,
      "author_name": "yuval6967",
      "author_url": "",
      "post_date": "06/17/2020 15:21:29",
      "content": "<p>As I understand these cell are malignant but not melanoma. When I used them for training I set the target to 0</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 890756,
      "author_name": "jensvannahl",
      "author_url": "",
      "post_date": "06/17/2020 17:18:44",
      "content": "<p>In the ISIC 2019 dataset are three different malignant skin lesions: Melanoma (MEL), basal cell carcinoma (BCC) and spinal cell carcinoma (SCC). \nAbout this, I am absolute sure. I work as medical doctor, deep learning is only a hobby for me.\nI think, Gilles Vandewiele is correct, that we only have to predict melanomas. But when not, our models forget to learn more than 8000 BCC an 800 SCC! This would make a big difference!</p>",
      "votes": null,
      "replies": [
        {
          "id": 890927,
          "author_name": "group16",
          "author_url": "",
          "post_date": "06/17/2020 19:10:55",
          "content": "<p>Is the symptomatology for BCC and SCC more similar to MEL than to benign lesions such as nevi? If that is the case, then adding those as negative examples could confuse the model indeed (especially if these are not present within the public/private test set)!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 891901,
      "author_name": "zzy990106",
      "author_url": "",
      "post_date": "06/18/2020 14:19:59",
      "content": "<p>Does ISIC 2020 contain basal cell carcinomas? </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 891990,
      "author_name": "veronicarotemberg",
      "author_url": "",
      "post_date": "06/18/2020 15:24:22",
      "content": "<p>Basal Cell Carcinomas and Squamous Cell Carcinomas are not considered Melanoma and would not be considered target = 1 for the purposes of this challenge. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 892290,
      "author_name": "jensvannahl",
      "author_url": "",
      "post_date": "06/18/2020 19:18:46",
      "content": "<p>Oh thank you very much for this clarification!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "888774": "External Datasets contain malignant images, which are different from melanomas: basal cell carcinoma and spinal cell carcinoma. When we use them, should the 'target' of the images only be '1' for the case of melanomas?\nIn all external dataset used here (from Shonenkow, Chris Deotte), the targets were set to '1' , when the image shows melanomas, but not in the case of basal cell carcinomas. \nThe ISIC 2019 dataset contains more than 8000 basal cell carcinomas, which are definitely malignant! \nThe question is, if the original dataset (and the test set) contains only melanomas as malignant images (because this is a Melanoma Classification Competition)!",
    "890483": "&gt;  basal cell carcinomas, which are definitely malignant\n\nThis is completely false. See the article. https://www.cancer.org/cancer/basal-and-squamous-cell-skin-cancer/about/what-is-basal-and-squamous-cell.html",
    "890575": "As I understand these cell are malignant but not melanoma. When I used them for training I set the target to 0",
    "890597": "carcinoma = a cancer arising in the epithelial tissue of the skin or of the lining of the internal organs.\n\ncancer = malignant tumor\n\nSo that statement is completely true. Nevertheless, this competition is about detecting melanomas as far as I know.",
    "890756": "In the ISIC 2019 dataset are three different malignant skin lesions: Melanoma (MEL), basal cell carcinoma (BCC) and spinal cell carcinoma (SCC). \nAbout this, I am absolute sure. I work as medical doctor, deep learning is only a hobby for me.\nI think, Gilles Vandewiele is correct, that we only have to predict melanomas. But when not, our models forget to learn more than 8000 BCC an 800 SCC! This would make a big difference!",
    "890927": "Is the symptomatology for BCC and SCC more similar to MEL than to benign lesions such as nevi? If that is the case, then adding those as negative examples could confuse the model indeed (especially if these are not present within the public/private test set)!",
    "891901": "Does ISIC 2020 contain basal cell carcinomas?",
    "891990": "Basal Cell Carcinomas and Squamous Cell Carcinomas are not considered Melanoma and would not be considered target = 1 for the purposes of this challenge.",
    "892290": "Oh thank you very much for this clarification!"
  },
  "source": "meta"
}