{
  "id": 381995,
  "title": "How to process data for multi view training?",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/381995",
  "author_name": "Yingshu Li",
  "post_date": "2023-01-29T04:11:43.036000",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I want to try using multi-view images to train the model. However, I found that some patients may have more than 4 images. I just wonder how to process this situation. If anyone has suggestions about how to process the dataset. Thanks a lot.</p>",
  "messages": [
    {
      "id": 2119734,
      "postDate": "2023-01-29T04:11:43.037Z",
      "content": "<p>I want to try using multi-view images to train the model. However, I found that some patients may have more than 4 images. I just wonder how to process this situation. If anyone has suggestions about how to process the dataset. Thanks a lot.</p>",
      "rawMarkdown": "I want to try using multi-view images to train the model. However, I found that some patients may have more than 4 images. I just wonder how to process this situation. If anyone has suggestions about how to process the dataset. Thanks a lot.",
      "votes": 3
    },
    {
      "id": 2127335,
      "postDate": "2023-02-02T20:14:05.710Z",
      "content": "<p>See my notebook: <a href=\"https://www.kaggle.com/code/anttiisosalo/solt-based-image-augs-rsna-bc-detection\" target=\"_blank\">🎗️ [SOLT-based image augs] RSNA BC Detection</a>. I limit to the 4 standard views, namely R-MLO, L-MLO, R-CC and L-CC. There can possibly be several images on some of those views and one way to choose is to take the one, which was imaged the last if timestamp is available. For examinations where some of the views are missing, it is perhaps possible to solve the issue with imputation.</p>",
      "rawMarkdown": "See my notebook: [🎗️ [SOLT-based image augs] RSNA BC Detection](https://www.kaggle.com/code/anttiisosalo/solt-based-image-augs-rsna-bc-detection). I limit to the 4 standard views, namely R-MLO, L-MLO, R-CC and L-CC. There can possibly be several images on some of those views and one way to choose is to take the one, which was imaged the last if timestamp is available. For examinations where some of the views are missing, it is perhaps possible to solve the issue with imputation."
    },
    {
      "id": 2120372,
      "postDate": "2023-01-29T14:56:12.763Z",
      "content": "<p>I use N view images to get N samples, and get their mean of prediction.</p>",
      "rawMarkdown": "I use N view images to get N samples, and get their mean of prediction."
    },
    {
      "id": 2119981,
      "postDate": "2023-01-29T09:15:41.130Z",
      "content": "<p>I randomly pick N pic from a patients.</p>",
      "rawMarkdown": "I randomly pick N pic from a patients."
    }
  ],
  "comments": [
    {
      "id": 2127335,
      "author_name": "Antti Isosalo",
      "author_url": "",
      "post_date": "2023-02-02T20:14:05.710000",
      "content": "<p>See my notebook: <a href=\"https://www.kaggle.com/code/anttiisosalo/solt-based-image-augs-rsna-bc-detection\" target=\"_blank\">🎗️ [SOLT-based image augs] RSNA BC Detection</a>. I limit to the 4 standard views, namely R-MLO, L-MLO, R-CC and L-CC. There can possibly be several images on some of those views and one way to choose is to take the one, which was imaged the last if timestamp is available. For examinations where some of the views are missing, it is perhaps possible to solve the issue with imputation.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2120372,
      "author_name": "yoyobar",
      "author_url": "",
      "post_date": "2023-01-29T14:56:12.763000",
      "content": "<p>I use N view images to get N samples, and get their mean of prediction.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2119981,
      "author_name": "Chenglu",
      "author_url": "",
      "post_date": "2023-01-29T09:15:41.130000",
      "content": "<p>I randomly pick N pic from a patients.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2119734": "I want to try using multi-view images to train the model. However, I found that some patients may have more than 4 images. I just wonder how to process this situation. If anyone has suggestions about how to process the dataset. Thanks a lot.",
    "2127335": "See my notebook: [🎗️ [SOLT-based image augs] RSNA BC Detection](https://www.kaggle.com/code/anttiisosalo/solt-based-image-augs-rsna-bc-detection). I limit to the 4 standard views, namely R-MLO, L-MLO, R-CC and L-CC. There can possibly be several images on some of those views and one way to choose is to take the one, which was imaged the last if timestamp is available. For examinations where some of the views are missing, it is perhaps possible to solve the issue with imputation.",
    "2120372": "I use N view images to get N samples, and get their mean of prediction.",
    "2119981": "I randomly pick N pic from a patients."
  }
}