{
  "id": 287713,
  "title": "3rd place solution ",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/287713",
  "author_name": "Cedric Soares",
  "post_date": "2021-11-15T10:49:13.871000",
  "votes": 13,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I would like to thank Kaggle, RSNA and MICCAI for hosting the competition. I would also like to thank <a href=\"https://www.kaggle.com/billqi\" target=\"_blank\">@billqi</a> for his kernel which I used as a work base for mine. </p>\n<h1>Approach</h1>\n<ul>\n<li>I used stratified split based on patient ids and class on train dataset to sample a validation dataset</li>\n<li>I trained four EfficientNet-B3. One for each kind of MRI scans (FLAIR, T1w, T1wCE, T2w) </li>\n<li>I aggregated results by patient ids to compare differences between the maximum prediction and the average of predictions to average average and the minimum of predictions to keep the prediction linked to the highest difference. </li>\n</ul>\n<h1>Notebooks</h1>\n<p>Notebook is available on Kaggle : <a href=\"https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split\" target=\"_blank\">https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split</a><br>\nAnd Github (with some comments on french) : <a href=\"https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification\" target=\"_blank\">https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification</a></p>",
  "messages": [
    {
      "id": 1582856,
      "postDate": "2021-11-15T10:49:13.873Z",
      "content": "<p>I would like to thank Kaggle, RSNA and MICCAI for hosting the competition. I would also like to thank <a href=\"https://www.kaggle.com/billqi\" target=\"_blank\">@billqi</a> for his kernel which I used as a work base for mine. </p>\n<h1>Approach</h1>\n<ul>\n<li>I used stratified split based on patient ids and class on train dataset to sample a validation dataset</li>\n<li>I trained four EfficientNet-B3. One for each kind of MRI scans (FLAIR, T1w, T1wCE, T2w) </li>\n<li>I aggregated results by patient ids to compare differences between the maximum prediction and the average of predictions to average average and the minimum of predictions to keep the prediction linked to the highest difference. </li>\n</ul>\n<h1>Notebooks</h1>\n<p>Notebook is available on Kaggle : <a href=\"https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split\" target=\"_blank\">https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split</a><br>\nAnd Github (with some comments on french) : <a href=\"https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification\" target=\"_blank\">https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification</a></p>",
      "rawMarkdown": "I would like to thank Kaggle, RSNA and MICCAI for hosting the competition. I would also like to thank @billqi for his kernel which I used as a work base for mine. \n\n# Approach \n- I used stratified split based on patient ids and class on train dataset to sample a validation dataset\n- I trained four EfficientNet-B3. One for each kind of MRI scans (FLAIR, T1w, T1wCE, T2w) \n- I aggregated results by patient ids to compare differences between the maximum prediction and the average of predictions to average average and the minimum of predictions to keep the prediction linked to the highest difference. \n\n# Notebooks\nNotebook is available on Kaggle : https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split\nAnd Github (with some comments on french) : https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification\n\n",
      "votes": 13
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1582856": "I would like to thank Kaggle, RSNA and MICCAI for hosting the competition. I would also like to thank @billqi for his kernel which I used as a work base for mine. \n\n# Approach \n- I used stratified split based on patient ids and class on train dataset to sample a validation dataset\n- I trained four EfficientNet-B3. One for each kind of MRI scans (FLAIR, T1w, T1wCE, T2w) \n- I aggregated results by patient ids to compare differences between the maximum prediction and the average of predictions to average average and the minimum of predictions to keep the prediction linked to the highest difference. \n\n# Notebooks\nNotebook is available on Kaggle : https://www.kaggle.com/cedricsoares/tf-efficientnet-transfer-learning-strat-split\nAnd Github (with some comments on french) : https://github.com/cedricsoares/kaggle-rsna-miccai-brain-tumor-radiogenomic-classification\n\n"
  }
}