Hello! Just saw that an older topic gives more informations about segmentation data. As I understood, TotalSegmentator model was fine tuned in order to segment only interesting organs for this competition. I have some questions :
\nThanks again I am open to discussion about it ๐๐ผ๐๐ผ
", "messages": [ { "id": 2378532, "postDate": "2023-08-07T17:16:36.893Z", "content": "The provided segmentation data on 206 scans were manually refined by radiologists after being segmented on a model trained on the total segmentator data with specific organs included. We didn't refine segmentations on all training cases due to the labeling effort required in revising voxelwise segmentations. You can train your own segmentation model from the total segmentator dataset and/or the 206 cases provided here.
", "rawMarkdown": "The provided segmentation data on 206 scans were manually refined by radiologists after being segmented on a model trained on the total segmentator data with specific organs included. We didn't refine segmentations on all training cases due to the labeling effort required in revising voxelwise segmentations. You can train your own segmentation model from the total segmentator dataset and/or the 206 cases provided here. ", "votes": 3, "replies": [ { "id": 2378555, "postDate": "2023-08-07T17:23:34.530Z", "content": "Thanks for you answer! Will try to train my own model ;)
", "rawMarkdown": "Thanks for you answer! Will try to train my own model ;) " }, { "id": 2411252, "postDate": "2023-08-27T13:59:26.437Z", "content": "Thanks for your answer.
\nAs we know, segmentation has not been provided for the whole specific series. I want to know how to pair an annotation instance with Dicom in a series?
My assumption is that we may take only the latest aortic-hu-valued series_id for each patient. Can anyone enlighten me regarding this?
", "rawMarkdown": "My assumption is that we may take only the latest aortic-hu-valued series_id for each patient. Can anyone enlighten me regarding this?" } ] } ] }, { "id": 2378291, "postDate": "2023-08-07T15:19:48.707Z", "content": "Hello! Just saw that an older topic gives more informations about segmentation data. As I understood, TotalSegmentator model was fine tuned in order to segment only interesting organs for this competition. I have some questions :
\nThanks again I am open to discussion about it ๐๐ผ๐๐ผ
", "rawMarkdown": "Hello! Just saw that an older topic gives more informations about segmentation data. As I understood, TotalSegmentator model was fine tuned in order to segment only interesting organs for this competition. I have some questions :\n- why donโt we have segmentation data for all scans ? Is it voluntary in order to reward people who succeed to use the model by their own? In particular I want to train expert models that are train to detect injuries for a particular organ so I want to give it frames that display the organ (thanks to segmentation data). Test data donโt segmentation data\n\nThanks again I am open to discussion about it ๐๐ผ๐๐ผ", "votes": 4 } ], "comments": [ { "id": 2378532, "author_name": "JeffRudie", "author_url": "", "post_date": "2023-08-07T17:16:36.893000", "content": "The provided segmentation data on 206 scans were manually refined by radiologists after being segmented on a model trained on the total segmentator data with specific organs included. We didn't refine segmentations on all training cases due to the labeling effort required in revising voxelwise segmentations. You can train your own segmentation model from the total segmentator dataset and/or the 206 cases provided here.
", "votes": 3, "replies": [ { "id": 2378555, "author_name": "Henri Upton", "author_url": "", "post_date": "2023-08-07T17:23:34.530000", "content": "Thanks for you answer! Will try to train my own model ;)
", "votes": 0, "replies": [] }, { "id": 2411252, "author_name": "Armin Azhdehnia", "author_url": "", "post_date": "2023-08-27T13:59:26.437000", "content": "Thanks for your answer.
\nAs we know, segmentation has not been provided for the whole specific series. I want to know how to pair an annotation instance with Dicom in a series?
My assumption is that we may take only the latest aortic-hu-valued series_id for each patient. Can anyone enlighten me regarding this?
", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2378532": "The provided segmentation data on 206 scans were manually refined by radiologists after being segmented on a model trained on the total segmentator data with specific organs included. We didn't refine segmentations on all training cases due to the labeling effort required in revising voxelwise segmentations. You can train your own segmentation model from the total segmentator dataset and/or the 206 cases provided here. ", "2378291": "Hello! Just saw that an older topic gives more informations about segmentation data. As I understood, TotalSegmentator model was fine tuned in order to segment only interesting organs for this competition. I have some questions :\n- why donโt we have segmentation data for all scans ? Is it voluntary in order to reward people who succeed to use the model by their own? In particular I want to train expert models that are train to detect injuries for a particular organ so I want to give it frames that display the organ (thanks to segmentation data). Test data donโt segmentation data\n\nThanks again I am open to discussion about it ๐๐ผ๐๐ผ" } }