Till now I've been blind resizing all the images inside the same folder to a common shape and respecting the instance ordering provided. But after a closer look to the series_id 3401861580 I'm not very sure that they're not just two different sets ordered and renamed so what we think is 1 2 3 4 5 6… actually is 1A 1B 2A 2B 3A 3B….
\nI'm not an expert but that would explain why the shapes are so mixed:
\narray([[640, 640],\n [640, 640],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [640, 640],\n [608, 608],
And why you see what I think are kidneys slices intercalated with aorta slices if you respect the instance number provided. What do you think?
\nEDIT : dcm.sort(key=lambda x:x.ImagePositionPatient[2]) fixed it, I've been tryng to avoid the use of the metadata and trust instance numbers provided but I think that at least for these cases is a must.
EDIT2: Could I use Covid-19 as excuse, but the truth is that on that example I'nt sorted by instance number, the inconsistencys described were caused for a plane glob reading.
", "messages": [ { "id": 2946119, "postDate": "2024-08-04T01:42:56.720Z", "content": "Till now I've been blind resizing all the images inside the same folder to a common shape and respecting the instance ordering provided. But after a closer look to the series_id 3401861580 I'm not very sure that they're not just two different sets ordered and renamed so what we think is 1 2 3 4 5 6… actually is 1A 1B 2A 2B 3A 3B….
\nI'm not an expert but that would explain why the shapes are so mixed:
\narray([[640, 640],\n [640, 640],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [640, 640],\n [608, 608],
And why you see what I think are kidneys slices intercalated with aorta slices if you respect the instance number provided. What do you think?
\nEDIT : dcm.sort(key=lambda x:x.ImagePositionPatient[2]) fixed it, I've been tryng to avoid the use of the metadata and trust instance numbers provided but I think that at least for these cases is a must.
EDIT2: Could I use Covid-19 as excuse, but the truth is that on that example I'nt sorted by instance number, the inconsistencys described were caused for a plane glob reading.
", "rawMarkdown": "Till now I've been blind resizing all the images inside the same folder to a common shape and respecting the instance ordering provided. But after a closer look to the series_id 3401861580 I'm not very sure that they're not just two different sets ordered and renamed so what we think is 1 2 3 4 5 6... actually is 1A 1B 2A 2B 3A 3B....\n\nI'm not an expert but that would explain why the shapes are so mixed:\n\n`array([[640, 640],\n [640, 640],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [640, 640],\n [608, 608],`\n\nAnd why you see what I think are kidneys slices intercalated with aorta slices if you respect the instance number provided. What do you think?\n\nEDIT : `dcm.sort(key=lambda x:x.ImagePositionPatient[2])` fixed it, I've been tryng to avoid the use of the metadata and trust instance numbers provided but I think that at least for these cases is a must.\n\nEDIT2: Could I use Covid-19 as excuse, but the truth is that on that example I'nt sorted by instance number, the inconsistencys described were caused for a plane glob reading.", "votes": 5 }, { "id": 2946303, "postDate": "2024-08-04T08:36:01.647Z", "content": "@sacuscreed
\nYou need to try and find what works best for you.
\nNeither metadata nor instance number from filename are reliable source of truth or sorting method for this competition.
See below Issues and Related Dicussions :
\nA. Issues with using file number
\nFile numbers are inconsistent, few files missing etc.
B. Issues with using metadata
\nmetadata is in sorted or reverse sorted order.
C. Issues with orientation
\nOrientation creates issues with sorting with patient position
I think this is becuase the data set is a combination of different sources.
\nso you can see cluster of \"different charcteristics\".
what i am afraid is that there are sources in the hidden test that is not found in the train.
\ne.g. would there be 1A 1B 2A 2B 3A 3B but size are the same? how about 1A 1B 1C 2A 2B 2C ?
", "rawMarkdown": "I think this is becuase the data set is a combination of different sources.\nso you can see cluster of \"different charcteristics\".\n\nwhat i am afraid is that there are sources in the hidden test that is not found in the train.\n\ne.g. would there be 1A 1B 2A 2B 3A 3B but size are the same? how about 1A 1B 1C 2A 2B 2C ?", "votes": 1, "replies": [ { "id": 2950736, "postDate": "2024-08-07T20:51:32.010Z", "content": "Sorry for the mistake. I was sick and In't payd attention. I checked again by correctly sorting by instance numbers and the situation described was not succeing. My fault.
\nI think that, reversed or not… , may be with some instances missing…, but instances numbers are correct since have been annotated by the same professionals than sorted and diagnosticated the volumes.
", "rawMarkdown": "Sorry for the mistake. I was sick and In't payd attention. I checked again by correctly sorting by instance numbers and the situation described was not succeing. My fault.\n\nI think that, reversed or not... , may be with some instances missing..., but instances numbers are correct since have been annotated by the same professionals than sorted and diagnosticated the volumes." } ] }, { "id": 2946747, "postDate": "2024-08-04T16:12:28.527Z", "content": "We are struggling with this as well. Without domain knowledge I find it difficult to determine the correctness of the order, orientation, instance numbers, etc.
", "rawMarkdown": "We are struggling with this as well. Without domain knowledge I find it difficult to determine the correctness of the order, orientation, instance numbers, etc.", "votes": 2 }, { "id": 2996033, "postDate": "2024-09-23T00:43:08.133Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 2946303, "author_name": "Rohit Chaudhari", "author_url": "", "post_date": "2024-08-04T08:36:01.647000", "content": "@sacuscreed
\nYou need to try and find what works best for you.
\nNeither metadata nor instance number from filename are reliable source of truth or sorting method for this competition.
See below Issues and Related Dicussions :
\nA. Issues with using file number
\nFile numbers are inconsistent, few files missing etc.
B. Issues with using metadata
\nmetadata is in sorted or reverse sorted order.
C. Issues with orientation
\nOrientation creates issues with sorting with patient position
I think this is becuase the data set is a combination of different sources.
\nso you can see cluster of \"different charcteristics\".
what i am afraid is that there are sources in the hidden test that is not found in the train.
\ne.g. would there be 1A 1B 2A 2B 3A 3B but size are the same? how about 1A 1B 1C 2A 2B 2C ?
", "votes": 1, "replies": [ { "id": 2950736, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2024-08-07T20:51:32.010000", "content": "Sorry for the mistake. I was sick and In't payd attention. I checked again by correctly sorting by instance numbers and the situation described was not succeing. My fault.
\nI think that, reversed or not… , may be with some instances missing…, but instances numbers are correct since have been annotated by the same professionals than sorted and diagnosticated the volumes.
", "votes": 0, "replies": [] } ] }, { "id": 2946747, "author_name": "Bartley", "author_url": "", "post_date": "2024-08-04T16:12:28.527000", "content": "We are struggling with this as well. Without domain knowledge I find it difficult to determine the correctness of the order, orientation, instance numbers, etc.
", "votes": 2, "replies": [] }, { "id": 2996033, "author_name": "", "author_url": "", "post_date": "2024-09-23T00:43:08.133000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2946119": "Till now I've been blind resizing all the images inside the same folder to a common shape and respecting the instance ordering provided. But after a closer look to the series_id 3401861580 I'm not very sure that they're not just two different sets ordered and renamed so what we think is 1 2 3 4 5 6... actually is 1A 1B 2A 2B 3A 3B....\n\nI'm not an expert but that would explain why the shapes are so mixed:\n\n`array([[640, 640],\n [640, 640],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [640, 640],\n [640, 640],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [608, 608],\n [640, 640],\n [608, 608],`\n\nAnd why you see what I think are kidneys slices intercalated with aorta slices if you respect the instance number provided. What do you think?\n\nEDIT : `dcm.sort(key=lambda x:x.ImagePositionPatient[2])` fixed it, I've been tryng to avoid the use of the metadata and trust instance numbers provided but I think that at least for these cases is a must.\n\nEDIT2: Could I use Covid-19 as excuse, but the truth is that on that example I'nt sorted by instance number, the inconsistencys described were caused for a plane glob reading.", "2946303": "@sacuscreed \nYou need to try and find what works best for you.\nNeither metadata nor instance number from filename are reliable source of truth or sorting method for this competition.\n\nSee below Issues and Related Dicussions : \n**A. Issues with using file number**\nFile numbers are inconsistent, few files missing etc.\n1. [inconsistency in instance number in file name](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/523550)\n2. [Inconsistent instance numbers for study_id = 2581283971 , series_id = 2683794967](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/519208)\n\n**B. Issues with using metadata**\nmetadata is in sorted or reverse sorted order.\n3. [Filenames vs metadata for slice image sorting](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/518525)\n\n**C. Issues with orientation**\nOrientation creates issues with sorting with patient position\n\n4. [2d to 3d coord conversion: Are the dicom tag wrong?](https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/522956)\n", "2950709": "I think this is becuase the data set is a combination of different sources.\nso you can see cluster of \"different charcteristics\".\n\nwhat i am afraid is that there are sources in the hidden test that is not found in the train.\n\ne.g. would there be 1A 1B 2A 2B 3A 3B but size are the same? how about 1A 1B 1C 2A 2B 2C ?", "2946747": "We are struggling with this as well. Without domain knowledge I find it difficult to determine the correctness of the order, orientation, instance numbers, etc.", "2996033": "" } }