{ "id": 436129, "title": "Learning preprocessing from nnU-Net?", "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/436129", "author_name": "RickyLu", "post_date": "2023-09-01T07:16:59.878000", "votes": 3, "comment_count": 2, "views": 0, "content": "

nnU-Net has implemented a detailed preprocessing pipeline which is suitable for the 3D data in this competition.

\n
    \n
  1. Crop: keep only the largest non-black rectangle area in the image.
    2. \n
  2. Resample: Find the median spacing of all the training images and resample all of them into a same spacing.
    3. \n
  3. Normalization: Choose 5%~95% of the grey scale value in a single image and normalize it.
    4. \n
\n

Paper: https://www.nature.com/articles/s41592-020-01008-z
\nGithub: https://github.com/MIC-DKFZ/nnUNet

", "messages": [ { "id": 2418160, "postDate": "2023-09-01T07:16:59.880Z", "content": "

nnU-Net has implemented a detailed preprocessing pipeline which is suitable for the 3D data in this competition.

\n
    \n
  1. Crop: keep only the largest non-black rectangle area in the image.
    2. \n
  2. Resample: Find the median spacing of all the training images and resample all of them into a same spacing.
    3. \n
  3. Normalization: Choose 5%~95% of the grey scale value in a single image and normalize it.
    4. \n
\n

Paper: https://www.nature.com/articles/s41592-020-01008-z
\nGithub: https://github.com/MIC-DKFZ/nnUNet

", "rawMarkdown": "nnU-Net has implemented a detailed preprocessing pipeline which is suitable for the 3D data in this competition.\n1. Crop: keep only the largest non-black rectangle area in the image.\n2. Resample: Find the median **spacing** of all the training images and resample all of them into a same spacing.\n3. Normalization: Choose 5%~95% of the grey scale value in a single image and normalize it.\n\nPaper: [https://www.nature.com/articles/s41592-020-01008-z](https://www.nature.com/articles/s41592-020-01008-z)\nGithub: [https://github.com/MIC-DKFZ/nnUNet](https://github.com/MIC-DKFZ/nnUNet)", "votes": 3 }, { "id": 2418168, "postDate": "2023-09-01T07:20:23.340Z", "content": "

Wow, this is fantastic! Thank you for sharing this valuable resource. Preprocessing is such a crucial step in medical image analysis, and nnU-Net's preprocessing pipeline seems comprehensive and well-thought-out. The idea of cropping to focus on the essential information, standardizing voxel spacing, and normalizing the grayscale values makes a lot of sense for better model performance. I'll definitely check out the paper and GitHub repository to learn more about the details of implementation. Thanks again for sharing this helpful information!

", "rawMarkdown": "Wow, this is fantastic! Thank you for sharing this valuable resource. Preprocessing is such a crucial step in medical image analysis, and nnU-Net's preprocessing pipeline seems comprehensive and well-thought-out. The idea of cropping to focus on the essential information, standardizing voxel spacing, and normalizing the grayscale values makes a lot of sense for better model performance. I'll definitely check out the paper and GitHub repository to learn more about the details of implementation. Thanks again for sharing this helpful information!\n", "votes": 1 }, { "id": 2418443, "postDate": "2023-09-01T09:25:19.737Z", "content": "

@rickylu thanks for sharing helpful resources.

", "rawMarkdown": "@rickylu thanks for sharing helpful resources." } ], "comments": [ { "id": 2418168, "author_name": "Oleg Zholobov", "author_url": "", "post_date": "2023-09-01T07:20:23.340000", "content": "

Wow, this is fantastic! Thank you for sharing this valuable resource. Preprocessing is such a crucial step in medical image analysis, and nnU-Net's preprocessing pipeline seems comprehensive and well-thought-out. The idea of cropping to focus on the essential information, standardizing voxel spacing, and normalizing the grayscale values makes a lot of sense for better model performance. I'll definitely check out the paper and GitHub repository to learn more about the details of implementation. Thanks again for sharing this helpful information!

", "votes": 1, "replies": [] }, { "id": 2418443, "author_name": "Faysal Miah", "author_url": "", "post_date": "2023-09-01T09:25:19.737000", "content": "

@rickylu thanks for sharing helpful resources.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2418160": "nnU-Net has implemented a detailed preprocessing pipeline which is suitable for the 3D data in this competition.\n1. Crop: keep only the largest non-black rectangle area in the image.\n2. Resample: Find the median **spacing** of all the training images and resample all of them into a same spacing.\n3. Normalization: Choose 5%~95% of the grey scale value in a single image and normalize it.\n\nPaper: [https://www.nature.com/articles/s41592-020-01008-z](https://www.nature.com/articles/s41592-020-01008-z)\nGithub: [https://github.com/MIC-DKFZ/nnUNet](https://github.com/MIC-DKFZ/nnUNet)", "2418168": "Wow, this is fantastic! Thank you for sharing this valuable resource. Preprocessing is such a crucial step in medical image analysis, and nnU-Net's preprocessing pipeline seems comprehensive and well-thought-out. The idea of cropping to focus on the essential information, standardizing voxel spacing, and normalizing the grayscale values makes a lot of sense for better model performance. I'll definitely check out the paper and GitHub repository to learn more about the details of implementation. Thanks again for sharing this helpful information!\n", "2418443": "@rickylu thanks for sharing helpful resources." } }