While reviewing the dataset descriptions, I noticed that patient identifiers are removed, meaning we are unable to determine if multiple scans come from the same patient or not.
\nI may have missed this, but do any patients contribute multiple scans to the dataset? If so, were any of these scans taken at the same time, or are they separated throughout time?
\nI want to know if any patients contributed multiple imaging modalities to the dataset, as that would have implications of being able to train a model to reconstruct one modality from another, or assist with cross-modality learning.
", "messages": [ { "id": 3259455, "postDate": "2025-08-01T13:39:12.053Z", "content": "While reviewing the dataset descriptions, I noticed that patient identifiers are removed, meaning we are unable to determine if multiple scans come from the same patient or not.
\nI may have missed this, but do any patients contribute multiple scans to the dataset? If so, were any of these scans taken at the same time, or are they separated throughout time?
\nI want to know if any patients contributed multiple imaging modalities to the dataset, as that would have implications of being able to train a model to reconstruct one modality from another, or assist with cross-modality learning.
", "rawMarkdown": "While reviewing the dataset descriptions, I noticed that patient identifiers are removed, meaning we are unable to determine if multiple scans come from the same patient or not. \n\nI may have missed this, but do any patients contribute multiple scans to the dataset? If so, were any of these scans taken at the same time, or are they separated throughout time?\n\nI want to know if any patients contributed multiple imaging modalities to the dataset, as that would have implications of being able to train a model to reconstruct one modality from another, or assist with cross-modality learning.", "votes": 7 }, { "id": 3266040, "postDate": "2025-08-08T09:37:48.353Z", "content": "this is also interesting with regard to “data leakage” for training and evaluation. If patients occur more than once, you should avoid distributing them across the different sets.
\nBut as Evan Calabrese has already written, these should be considered independent. I just wanted to explicitly mention this type of issue once.
https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/597070
\nadi_hatake mentioned that already. sorry I didn't read that before posting.
We are not providing any further info on this at this time. Please treat series as independent.
", "rawMarkdown": "We are not providing any further info on this at this time. Please treat series as independent. " } ], "comments": [ { "id": 3266040, "author_name": "Nicolai Krekiehn", "author_url": "", "post_date": "2025-08-08T09:37:48.353000", "content": "this is also interesting with regard to “data leakage” for training and evaluation. If patients occur more than once, you should avoid distributing them across the different sets.
\nBut as Evan Calabrese has already written, these should be considered independent. I just wanted to explicitly mention this type of issue once.
https://www.kaggle.com/competitions/rsna-intracranial-aneurysm-detection/discussion/597070
\nadi_hatake mentioned that already. sorry I didn't read that before posting.
We are not providing any further info on this at this time. Please treat series as independent.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3259455": "While reviewing the dataset descriptions, I noticed that patient identifiers are removed, meaning we are unable to determine if multiple scans come from the same patient or not. \n\nI may have missed this, but do any patients contribute multiple scans to the dataset? If so, were any of these scans taken at the same time, or are they separated throughout time?\n\nI want to know if any patients contributed multiple imaging modalities to the dataset, as that would have implications of being able to train a model to reconstruct one modality from another, or assist with cross-modality learning.", "3266040": "this is also interesting with regard to “data leakage” for training and evaluation. If patients occur more than once, you should avoid distributing them across the different sets.\nBut as Evan Calabrese has already written, these should be considered independent. I just wanted to explicitly mention this type of issue once.", "3265696": "We are not providing any further info on this at this time. Please treat series as independent. " } }