Hi all. This is my first time participating in such competitions. I have a question regarding data. Why is only bowel injury shown in the labels file? Where are the injuries to other organs? Or have I misunderstood something?
", "messages": [ { "id": 2365589, "postDate": "2023-07-30T11:00:30.160Z", "content": "Hi all. This is my first time participating in such competitions. I have a question regarding data. Why is only bowel injury shown in the labels file? Where are the injuries to other organs? Or have I misunderstood something?
", "rawMarkdown": "Hi all. This is my first time participating in such competitions. I have a question regarding data. Why is only bowel injury shown in the labels file? Where are the injuries to other organs? Or have I misunderstood something?", "votes": 3 }, { "id": 2366452, "postDate": "2023-07-31T02:50:13.277Z", "content": "My guess is that image level labels only gives the injury types that are visible in certain slices, while other injury types such as liver are visible across all slices
", "rawMarkdown": "My guess is that image level labels only gives the injury types that are visible in certain slices, while other injury types such as liver are visible across all slices", "votes": 1 }, { "id": 2365960, "postDate": "2023-07-30T16:14:12.350Z", "content": "As I understand it, image_level_labels.csv gives extra information, those aren't the target values. You are predicting the values in train.csv.
", "rawMarkdown": "As I understand it, image_level_labels.csv gives extra information, those aren't the target values. You are predicting the values in train.csv. ", "votes": 1, "replies": [ { "id": 2365986, "postDate": "2023-07-30T16:33:50.707Z", "content": "i know that, bro, but in train.csv each example is just one patient id and each patient has one or more series with many single images. Therefore, should all individual patient images have the same label? As far as I know, only some images (slices) show the presence of organs or signs of damage, so we cannot look at an image without the liver and say the liver is diseased. Do you know what I mean?
", "rawMarkdown": "i know that, bro, but in train.csv each example is just one patient id and each patient has one or more series with many single images. Therefore, should all individual patient images have the same label? As far as I know, only some images (slices) show the presence of organs or signs of damage, so we cannot look at an image without the liver and say the liver is diseased. Do you know what I mean?", "replies": [ { "id": 2366172, "postDate": "2023-07-30T19:27:51.663Z", "content": "Yeah I know what you mean. I've been thinking about that as well, there's got to be some way to limit the slices with an algorithm or something to just the ones that show damage / are relevant to a prediction. Otherwise yeah idk another way around it besides labeling all of them. I was gonna go through a few of them manually and plot them to see where the cutoff should be, because I definitely saw some images of straight up legs, which obviously isn't useful for determining kidney health.
", "rawMarkdown": "Yeah I know what you mean. I've been thinking about that as well, there's got to be some way to limit the slices with an algorithm or something to just the ones that show damage / are relevant to a prediction. Otherwise yeah idk another way around it besides labeling all of them. I was gonna go through a few of them manually and plot them to see where the cutoff should be, because I definitely saw some images of straight up legs, which obviously isn't useful for determining kidney health." } ] }, { "id": 2367415, "postDate": "2023-07-31T14:41:49.707Z", "content": "Ok, then it turns out that we have a list of 15 values in the form of a single label? This is if the train.csv file is used as labels. I understand that we use the data in the series of images for training, but these images are not sorted by organ names. How do I tell the program, for example, a diseased liver or a healthy liver? Weird dataset.
", "rawMarkdown": "Ok, then it turns out that we have a list of 15 values in the form of a single label? This is if the train.csv file is used as labels. I understand that we use the data in the series of images for training, but these images are not sorted by organ names. How do I tell the program, for example, a diseased liver or a healthy liver? Weird dataset.", "replies": [ { "id": 2367659, "postDate": "2023-07-31T17:29:25.330Z", "content": "This is something I've also been thinking about. There's other posts in the discussion forum showing solutions from previous competitions, I still need to look through a few more of those. A possibly strategy is to do an ensemble of models, one for each column in train.csv, but that may be more difficult than expected due to the limitations on space and the size of the data.
", "rawMarkdown": "This is something I've also been thinking about. There's other posts in the discussion forum showing solutions from previous competitions, I still need to look through a few more of those. A possibly strategy is to do an ensemble of models, one for each column in train.csv, but that may be more difficult than expected due to the limitations on space and the size of the data." }, { "id": 2368627, "postDate": "2023-08-01T09:18:52.217Z", "content": "I have one guess. Maybe, there is no different on what organ we are looking at. Maybe the trauma's features look the same on every organ. And the bowel's trauma is enough to train model and then we will apply it to any organs.
", "rawMarkdown": "I have one guess. Maybe, there is no different on what organ we are looking at. Maybe the trauma's features look the same on every organ. And the bowel's trauma is enough to train model and then we will apply it to any organs. " } ] } ] }, { "id": 2369514, "postDate": "2023-08-01T19:33:27.440Z", "content": "We have clarified the labels provided on this post.
\nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/428538
The segmentations should provide enough context for narrowing the location to search for injuries in the liver, kidneys and spleen. However, extravasation and bowel injury have more variable anatomic locations and appearances. Therefore, image-level labels were also generated for these.
", "rawMarkdown": "We have clarified the labels provided on this post. \nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/428538\n\nThe segmentations should provide enough context for narrowing the location to search for injuries in the liver, kidneys and spleen. However, extravasation and bowel injury have more variable anatomic locations and appearances. Therefore, image-level labels were also generated for these. " }, { "id": 2365763, "postDate": "2023-07-30T13:08:01.807Z", "content": "Me too. There are 5 organs but image_level_labels has just 2 (Bowel and Active_Extravasation).
", "rawMarkdown": "Me too. There are 5 organs but image_level_labels has just 2 (Bowel and Active_Extravasation)." } ], "comments": [ { "id": 2366452, "author_name": "Feng Qilong", "author_url": "", "post_date": "2023-07-31T02:50:13.277000", "content": "My guess is that image level labels only gives the injury types that are visible in certain slices, while other injury types such as liver are visible across all slices
", "votes": 1, "replies": [] }, { "id": 2365960, "author_name": "Art. Berz.", "author_url": "", "post_date": "2023-07-30T16:14:12.350000", "content": "As I understand it, image_level_labels.csv gives extra information, those aren't the target values. You are predicting the values in train.csv.
", "votes": 1, "replies": [ { "id": 2365986, "author_name": "Vietnamese-NHNAM", "author_url": "", "post_date": "2023-07-30T16:33:50.707000", "content": "i know that, bro, but in train.csv each example is just one patient id and each patient has one or more series with many single images. Therefore, should all individual patient images have the same label? As far as I know, only some images (slices) show the presence of organs or signs of damage, so we cannot look at an image without the liver and say the liver is diseased. Do you know what I mean?
", "votes": 0, "replies": [ { "id": 2366172, "author_name": "Art. Berz.", "author_url": "", "post_date": "2023-07-30T19:27:51.663000", "content": "Yeah I know what you mean. I've been thinking about that as well, there's got to be some way to limit the slices with an algorithm or something to just the ones that show damage / are relevant to a prediction. Otherwise yeah idk another way around it besides labeling all of them. I was gonna go through a few of them manually and plot them to see where the cutoff should be, because I definitely saw some images of straight up legs, which obviously isn't useful for determining kidney health.
", "votes": 0, "replies": [] } ] }, { "id": 2367415, "author_name": "Oleg Zadneprovskyi", "author_url": "", "post_date": "2023-07-31T14:41:49.707000", "content": "Ok, then it turns out that we have a list of 15 values in the form of a single label? This is if the train.csv file is used as labels. I understand that we use the data in the series of images for training, but these images are not sorted by organ names. How do I tell the program, for example, a diseased liver or a healthy liver? Weird dataset.
", "votes": 0, "replies": [ { "id": 2367659, "author_name": "Art. Berz.", "author_url": "", "post_date": "2023-07-31T17:29:25.330000", "content": "This is something I've also been thinking about. There's other posts in the discussion forum showing solutions from previous competitions, I still need to look through a few more of those. A possibly strategy is to do an ensemble of models, one for each column in train.csv, but that may be more difficult than expected due to the limitations on space and the size of the data.
", "votes": 0, "replies": [] }, { "id": 2368627, "author_name": "Oleg Zadneprovskyi", "author_url": "", "post_date": "2023-08-01T09:18:52.217000", "content": "I have one guess. Maybe, there is no different on what organ we are looking at. Maybe the trauma's features look the same on every organ. And the bowel's trauma is enough to train model and then we will apply it to any organs.
", "votes": 0, "replies": [] } ] } ] }, { "id": 2369514, "author_name": "JeffRudie", "author_url": "", "post_date": "2023-08-01T19:33:27.440000", "content": "We have clarified the labels provided on this post.
\nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/428538
The segmentations should provide enough context for narrowing the location to search for injuries in the liver, kidneys and spleen. However, extravasation and bowel injury have more variable anatomic locations and appearances. Therefore, image-level labels were also generated for these.
", "votes": 0, "replies": [] }, { "id": 2365763, "author_name": "Vietnamese-NHNAM", "author_url": "", "post_date": "2023-07-30T13:08:01.807000", "content": "Me too. There are 5 organs but image_level_labels has just 2 (Bowel and Active_Extravasation).
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2365589": "Hi all. This is my first time participating in such competitions. I have a question regarding data. Why is only bowel injury shown in the labels file? Where are the injuries to other organs? Or have I misunderstood something?", "2366452": "My guess is that image level labels only gives the injury types that are visible in certain slices, while other injury types such as liver are visible across all slices", "2365960": "As I understand it, image_level_labels.csv gives extra information, those aren't the target values. You are predicting the values in train.csv. ", "2369514": "We have clarified the labels provided on this post. \nhttps://www.kaggle.com/competitions/rsna-2023-abdominal-trauma-detection/discussion/428538\n\nThe segmentations should provide enough context for narrowing the location to search for injuries in the liver, kidneys and spleen. However, extravasation and bowel injury have more variable anatomic locations and appearances. Therefore, image-level labels were also generated for these. ", "2365763": "Me too. There are 5 organs but image_level_labels has just 2 (Bowel and Active_Extravasation)." } }