{ "id": 444399, "title": "How can I create training data for the semantic segmentation model?", "url": "/competitions/rsna-2023-abdominal-trauma-detection/discussion/444399", "author_name": "Ataracsia", "post_date": "2023-10-01T19:13:21.682000", "votes": 2, "comment_count": 3, "views": 0, "content": "

The overview of this competition states the following as a description of segmentations:

\n
\n

segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set.

\n
\n

I can see a series of masked CT images and meta-information in the .nii file in segmentations, but I do not know how to use this to create training data for the Semantic Segmentation Model.

\n

For example, in the 397.nii file I get the following images and information, but how can I use this to annotate which organ is which color?

\n

Image 31 of 397.nii:
\n\"\"

\n
<class '>\ndata shape (, , )\naffine:\n[[  -   -.            .           ]\n [  -.               -.         -]\n [   .            .                    -]\n [   .            .            .            .        ]]\nmetadata:\n<class '> object, endian=\nsizeof_hdr      : 348\ndata_type       : ''\ndb_name         : ''\nextents         : 16384\nsession_error   : 0\nregular         : \ndim_info        : 0\ndim             : [  3 512 512  64   1   1   1   1]\nintent_p1       : 0.0\nintent_p2       : 0.0\nintent_p3       : 0.0\nintent_code     : \ndatatype        : \nbitpix          : 32\nslice_start     : 0\npixdim          : [-1.        0.904297  0.904297  2.5       0.        0.        0.\n  .      ]\nvox_offset      : 0.0\nscl_slope       : \nscl_inter       : \nslice_end       : 0\nslice_code      : \nxyzt_units      : 10\ncal_max         : 0.0\ncal_min         : 0.0\nslice_duration  : 0.0\ntoffset         : 0.0\nglmax           : 0\nglmin           : 0\ndescrip         : '-dirty --T15::+:'\naux_file        : ''\nqform_code      : \nsform_code      : \nquatern_b       : 0.0\nquatern_c       : 1.0\nquatern_d       : 0.0\nqoffset_x       : 232.252\nqoffset_y       : -223.81377\nqoffset_z       : -315.199\nsrow_x          : [ -0.904297  -0.         0.       232.252   ]\nsrow_y          : [  -0.          0.904297   -0.       -223.81377 ]\nsrow_z          : [   0.       0.       2.5   -315.199]\nintent_name     : ''\nmagic           : +'\n
", "messages": [ { "id": 2463983, "postDate": "2023-10-01T19:13:21.683Z", "content": "

The overview of this competition states the following as a description of segmentations:

\n
\n

segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set.

\n
\n

I can see a series of masked CT images and meta-information in the .nii file in segmentations, but I do not know how to use this to create training data for the Semantic Segmentation Model.

\n

For example, in the 397.nii file I get the following images and information, but how can I use this to annotate which organ is which color?

\n

Image 31 of 397.nii:
\n\"\"

\n
<class '>\ndata shape (, , )\naffine:\n[[  -   -.            .           ]\n [  -.               -.         -]\n [   .            .                    -]\n [   .            .            .            .        ]]\nmetadata:\n<class '> object, endian=\nsizeof_hdr      : 348\ndata_type       : ''\ndb_name         : ''\nextents         : 16384\nsession_error   : 0\nregular         : \ndim_info        : 0\ndim             : [  3 512 512  64   1   1   1   1]\nintent_p1       : 0.0\nintent_p2       : 0.0\nintent_p3       : 0.0\nintent_code     : \ndatatype        : \nbitpix          : 32\nslice_start     : 0\npixdim          : [-1.        0.904297  0.904297  2.5       0.        0.        0.\n  .      ]\nvox_offset      : 0.0\nscl_slope       : \nscl_inter       : \nslice_end       : 0\nslice_code      : \nxyzt_units      : 10\ncal_max         : 0.0\ncal_min         : 0.0\nslice_duration  : 0.0\ntoffset         : 0.0\nglmax           : 0\nglmin           : 0\ndescrip         : '-dirty --T15::+:'\naux_file        : ''\nqform_code      : \nsform_code      : \nquatern_b       : 0.0\nquatern_c       : 1.0\nquatern_d       : 0.0\nqoffset_x       : 232.252\nqoffset_y       : -223.81377\nqoffset_z       : -315.199\nsrow_x          : [ -0.904297  -0.         0.       232.252   ]\nsrow_y          : [  -0.          0.904297   -0.       -223.81377 ]\nsrow_z          : [   0.       0.       2.5   -315.199]\nintent_name     : ''\nmagic           : +'\n
", "rawMarkdown": "The overview of this competition states the following as a description of `segmentations`:\n\n>segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set.\n\nI can see a series of masked CT images and meta-information in the .nii file in `segmentations`, but I do not know how to use this to create training data for the Semantic Segmentation Model.\n\nFor example, in the 397.nii file I get the following images and information, but how can I use this to annotate which organ is which color?\n\nImage 31 of 397.nii:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F7389764%2Fc598cd637b107cd5062e3ecad5c309aa%2Foutput.png?generation=1696187248406329&alt=media)\n\n```\n\ndata shape (512, 512, 64)\naffine:\n[[ -0.90429699 -0. 0. 232.2519989 ]\n [ -0. 0.90429699 -0. -223.81376648]\n [ 0. 0. 2.5 -315.19900513]\n [ 0. 0. 0. 1. ]]\nmetadata:\n object, endian='<'\nsizeof_hdr : 348\ndata_type : b''\ndb_name : b''\nextents : 16384\nsession_error : 0\nregular : b'r'\ndim_info : 0\ndim : [ 3 512 512 64 1 1 1 1]\nintent_p1 : 0.0\nintent_p2 : 0.0\nintent_p3 : 0.0\nintent_code : none\ndatatype : float32\nbitpix : 32\nslice_start : 0\npixdim : [-1. 0.904297 0.904297 2.5 0. 0. 0.\n 0. ]\nvox_offset : 0.0\nscl_slope : nan\nscl_inter : nan\nslice_end : 0\nslice_code : unknown\nxyzt_units : 10\ncal_max : 0.0\ncal_min : 0.0\nslice_duration : 0.0\ntoffset : 0.0\nglmax : 0\nglmin : 0\ndescrip : b'2203.6-dirty 2022-09-08T15:38:56+01:00'\naux_file : b''\nqform_code : scanner\nsform_code : scanner\nquatern_b : 0.0\nquatern_c : 1.0\nquatern_d : 0.0\nqoffset_x : 232.252\nqoffset_y : -223.81377\nqoffset_z : -315.199\nsrow_x : [ -0.904297 -0. 0. 232.252 ]\nsrow_y : [ -0. 0.904297 -0. -223.81377 ]\nsrow_z : [ 0. 0. 2.5 -315.199]\nintent_name : b''\nmagic : b'n+1'\n```", "votes": 2 }, { "id": 2473061, "postDate": "2023-10-07T20:43:39.353Z", "content": "

segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set. This data is provided in the nifti file format. The filenames are series IDs. You can find a description of the source model (total segmentator) here and the data used to train that model here.

\n

Here you can find a mapping of the TotalSegmentator classes to SNOMED-CT codes.

", "rawMarkdown": "segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set. This data is provided in the nifti file format. The filenames are series IDs. You can find a description of the source model (total segmentator) [here ](https://pubs.rsna.org/doi/10.1148/ryai.230024)and the data used to train that model [here](https://github.com/wasserth/TotalSegmentator).\n\n[Here](https://github.com/wasserth/TotalSegmentator/blob/master/resources/totalsegmentator_snomed_mapping.csv) you can find a mapping of the TotalSegmentator classes to SNOMED-CT codes." }, { "id": 2473044, "postDate": "2023-10-07T20:12:49.337Z", "content": "

Hello. I think those masks should be usefull although you couldn't tell what organs they belong to.

", "rawMarkdown": "Hello. I think those masks should be usefull although you couldn't tell what organs they belong to. ", "replies": [ { "id": 2475103, "postDate": "2023-10-09T16:30:36.223Z", "content": "

Thank you. I understood it when I looked at the pixel values instead of looking at it as an image.

", "rawMarkdown": "Thank you. I understood it when I looked at the pixel values instead of looking at it as an image." } ] } ], "comments": [ { "id": 2473061, "author_name": "Ruy Diaz", "author_url": "", "post_date": "2023-10-07T20:43:39.353000", "content": "

segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set. This data is provided in the nifti file format. The filenames are series IDs. You can find a description of the source model (total segmentator) here and the data used to train that model here.

\n

Here you can find a mapping of the TotalSegmentator classes to SNOMED-CT codes.

", "votes": 0, "replies": [] }, { "id": 2473044, "author_name": "Ruy Diaz", "author_url": "", "post_date": "2023-10-07T20:12:49.337000", "content": "

Hello. I think those masks should be usefull although you couldn't tell what organs they belong to.

", "votes": 0, "replies": [ { "id": 2475103, "author_name": "Ataracsia", "author_url": "", "post_date": "2023-10-09T16:30:36.223000", "content": "

Thank you. I understood it when I looked at the pixel values instead of looking at it as an image.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "2463983": "The overview of this competition states the following as a description of `segmentations`:\n\n>segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set.\n\nI can see a series of masked CT images and meta-information in the .nii file in `segmentations`, but I do not know how to use this to create training data for the Semantic Segmentation Model.\n\nFor example, in the 397.nii file I get the following images and information, but how can I use this to annotate which organ is which color?\n\nImage 31 of 397.nii:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F7389764%2Fc598cd637b107cd5062e3ecad5c309aa%2Foutput.png?generation=1696187248406329&alt=media)\n\n```\n\ndata shape (512, 512, 64)\naffine:\n[[ -0.90429699 -0. 0. 232.2519989 ]\n [ -0. 0.90429699 -0. -223.81376648]\n [ 0. 0. 2.5 -315.19900513]\n [ 0. 0. 0. 1. ]]\nmetadata:\n object, endian='<'\nsizeof_hdr : 348\ndata_type : b''\ndb_name : b''\nextents : 16384\nsession_error : 0\nregular : b'r'\ndim_info : 0\ndim : [ 3 512 512 64 1 1 1 1]\nintent_p1 : 0.0\nintent_p2 : 0.0\nintent_p3 : 0.0\nintent_code : none\ndatatype : float32\nbitpix : 32\nslice_start : 0\npixdim : [-1. 0.904297 0.904297 2.5 0. 0. 0.\n 0. ]\nvox_offset : 0.0\nscl_slope : nan\nscl_inter : nan\nslice_end : 0\nslice_code : unknown\nxyzt_units : 10\ncal_max : 0.0\ncal_min : 0.0\nslice_duration : 0.0\ntoffset : 0.0\nglmax : 0\nglmin : 0\ndescrip : b'2203.6-dirty 2022-09-08T15:38:56+01:00'\naux_file : b''\nqform_code : scanner\nsform_code : scanner\nquatern_b : 0.0\nquatern_c : 1.0\nquatern_d : 0.0\nqoffset_x : 232.252\nqoffset_y : -223.81377\nqoffset_z : -315.199\nsrow_x : [ -0.904297 -0. 0. 232.252 ]\nsrow_y : [ -0. 0.904297 -0. -223.81377 ]\nsrow_z : [ 0. 0. 2.5 -315.199]\nintent_name : b''\nmagic : b'n+1'\n```", "2473061": "segmentations/ Model generated pixel-level annotations of the relevant organs and some major bones for a subset of the scans in the training set. This data is provided in the nifti file format. The filenames are series IDs. You can find a description of the source model (total segmentator) [here ](https://pubs.rsna.org/doi/10.1148/ryai.230024)and the data used to train that model [here](https://github.com/wasserth/TotalSegmentator).\n\n[Here](https://github.com/wasserth/TotalSegmentator/blob/master/resources/totalsegmentator_snomed_mapping.csv) you can find a mapping of the TotalSegmentator classes to SNOMED-CT codes.", "2473044": "Hello. I think those masks should be usefull although you couldn't tell what organs they belong to. " } }