{
  "id": 242070,
  "title": "APTOS-2019 Data Acquisition Condition",
  "url": "/competitions/aptos2019-blindness-detection/discussion/242070",
  "author_name": "Ammar Chalifah",
  "post_date": "2021-05-27T12:02:28.651000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>After working with APTOS-2019 dataset, I found several things that concerning.</p>\n<p>1) Based on diabetic retinopathy classification guide (I use ICDR as the standard), there are some sets of rules regarding the stage of severity based on visual cues. For example, mild DR has microaneurysm, but no exudates. However, there are some cases in this dataset where the image's visual features do not match the given label. For example, several mild DRs has many hard exudates.<br>\n2) The images have different aspect ratio, size, lighting, etc. For example, most of the normal images have full retina image, however the mild DR have clipped retina (the image not showing full retina).</p>\n<p>Based on that finding, I want to find out the dataset image acquisition process, i.e.<br>\na) from which hospital(s) are the data acquired from?<br>\nb) what is the method used for the labelling?<br>\nc) how many ophthalmologist/doctors/experts are being involved with the labelling process?</p>\n<p>If any of you guys know these information, or where I can get these info from, please let me know. Thank you!</p>",
  "messages": [
    {
      "id": 1324981,
      "postDate": "2021-05-27T12:02:28.650Z",
      "content": "<p>After working with APTOS-2019 dataset, I found several things that concerning.</p>\n<p>1) Based on diabetic retinopathy classification guide (I use ICDR as the standard), there are some sets of rules regarding the stage of severity based on visual cues. For example, mild DR has microaneurysm, but no exudates. However, there are some cases in this dataset where the image's visual features do not match the given label. For example, several mild DRs has many hard exudates.<br>\n2) The images have different aspect ratio, size, lighting, etc. For example, most of the normal images have full retina image, however the mild DR have clipped retina (the image not showing full retina).</p>\n<p>Based on that finding, I want to find out the dataset image acquisition process, i.e.<br>\na) from which hospital(s) are the data acquired from?<br>\nb) what is the method used for the labelling?<br>\nc) how many ophthalmologist/doctors/experts are being involved with the labelling process?</p>\n<p>If any of you guys know these information, or where I can get these info from, please let me know. Thank you!</p>",
      "rawMarkdown": "After working with APTOS-2019 dataset, I found several things that concerning.\n\n1) Based on diabetic retinopathy classification guide (I use ICDR as the standard), there are some sets of rules regarding the stage of severity based on visual cues. For example, mild DR has microaneurysm, but no exudates. However, there are some cases in this dataset where the image's visual features do not match the given label. For example, several mild DRs has many hard exudates.\n2) The images have different aspect ratio, size, lighting, etc. For example, most of the normal images have full retina image, however the mild DR have clipped retina (the image not showing full retina).\n\nBased on that finding, I want to find out the dataset image acquisition process, i.e.\na) from which hospital(s) are the data acquired from?\nb) what is the method used for the labelling?\nc) how many ophthalmologist/doctors/experts are being involved with the labelling process?\n\nIf any of you guys know these information, or where I can get these info from, please let me know. Thank you!"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1324981": "After working with APTOS-2019 dataset, I found several things that concerning.\n\n1) Based on diabetic retinopathy classification guide (I use ICDR as the standard), there are some sets of rules regarding the stage of severity based on visual cues. For example, mild DR has microaneurysm, but no exudates. However, there are some cases in this dataset where the image's visual features do not match the given label. For example, several mild DRs has many hard exudates.\n2) The images have different aspect ratio, size, lighting, etc. For example, most of the normal images have full retina image, however the mild DR have clipped retina (the image not showing full retina).\n\nBased on that finding, I want to find out the dataset image acquisition process, i.e.\na) from which hospital(s) are the data acquired from?\nb) what is the method used for the labelling?\nc) how many ophthalmologist/doctors/experts are being involved with the labelling process?\n\nIf any of you guys know these information, or where I can get these info from, please let me know. Thank you!"
  }
}