{
  "id": 183125,
  "title": "Study level and Image/Slice level prediction using single model?",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/183125",
  "author_name": "Abdur Rehman",
  "post_date": "2020-09-15T15:51:27.786000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Can we use single model for predicting both study and image/slice level prediction? My thought process right now is to predict 10 labels (9 study/exam level label + 1 image level label) and then for submission consider only 1 image level label for each image and for study level label, take mean/max of remaining 9 labels(since each study will now have multiple label rows). I am still very confused about study level prediction. I am opening this discussion for the expert suggestions which will be much appreciated.</p>\n<p>Thanks,</p>",
  "messages": [
    {
      "id": 1011639,
      "postDate": "2020-09-15T15:51:27.787Z",
      "content": "<p>Can we use single model for predicting both study and image/slice level prediction? My thought process right now is to predict 10 labels (9 study/exam level label + 1 image level label) and then for submission consider only 1 image level label for each image and for study level label, take mean/max of remaining 9 labels(since each study will now have multiple label rows). I am still very confused about study level prediction. I am opening this discussion for the expert suggestions which will be much appreciated.</p>\n<p>Thanks,</p>",
      "rawMarkdown": "Can we use single model for predicting both study and image/slice level prediction? My thought process right now is to predict 10 labels (9 study/exam level label + 1 image level label) and then for submission consider only 1 image level label for each image and for study level label, take mean/max of remaining 9 labels(since each study will now have multiple label rows). I am still very confused about study level prediction. I am opening this discussion for the expert suggestions which will be much appreciated.\n\nThanks,",
      "votes": 2
    },
    {
      "id": 1036455,
      "postDate": "2020-10-03T19:08:58.053Z",
      "content": "<p>Hi Abdur, I am having the same doubt. Let me know if you went with the solution you suggested, or if you did something else. Thanks.</p>",
      "rawMarkdown": "Hi Abdur, I am having the same doubt. Let me know if you went with the solution you suggested, or if you did something else. Thanks."
    }
  ],
  "comments": [
    {
      "id": 1036455,
      "author_name": "Jayasuriya Senthilvelan",
      "author_url": "",
      "post_date": "2020-10-03T19:08:58.053000",
      "content": "<p>Hi Abdur, I am having the same doubt. Let me know if you went with the solution you suggested, or if you did something else. Thanks.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1011639": "Can we use single model for predicting both study and image/slice level prediction? My thought process right now is to predict 10 labels (9 study/exam level label + 1 image level label) and then for submission consider only 1 image level label for each image and for study level label, take mean/max of remaining 9 labels(since each study will now have multiple label rows). I am still very confused about study level prediction. I am opening this discussion for the expert suggestions which will be much appreciated.\n\nThanks,",
    "1036455": "Hi Abdur, I am having the same doubt. Let me know if you went with the solution you suggested, or if you did something else. Thanks."
  }
}