From the paper describing the methods to create the dataset, I see that \"The collected raw data was mostly of adult PA-view CXRs, but also included a significant amount of outliers such as images of body parts other than chest (due to mismatched DICOM tags), pediatric scans, low-quality images, or lateral CXRs. […] All outliers were automatically excluded from the dataset […].\"
\nWhen I did some exploratory data analysis, I saw 839 .dicom files with PatientAge given as 000Y (plus 000D = 0 days?), which might either be yet another placeholder for missing or might really mean 0 years of age (i.e. < 1 year-old). I'm thinking that some of these might really be that, since some of the ages are below < 18 years, e.g. there's 107 images with 001Y to 017Y.
I'm curious on whether there's more information in this, because obviously some diagnoses are just much more likely at certain ages.
", "messages": [ { "id": 1136780, "postDate": "2021-01-03T12:25:02.173Z", "content": "I'm one of the papers' authors and yes, this subset is adult-only.
", "rawMarkdown": "I'm one of the papers' authors and yes, this subset is adult-only.", "votes": 3, "replies": [ { "id": 1136784, "postDate": "2021-01-03T12:28:41.193Z", "content": "Thanks, so stuff like Patient age ='012Y' in the .dicom meta-data would be a data issue?
", "rawMarkdown": "Thanks, so stuff like Patient age ='012Y' in the .dicom meta-data would be a data issue?", "votes": 1 } ] }, { "id": 1135183, "postDate": "2021-01-02T00:58:03.643Z", "content": "From the paper describing the methods to create the dataset, I see that \"The collected raw data was mostly of adult PA-view CXRs, but also included a significant amount of outliers such as images of body parts other than chest (due to mismatched DICOM tags), pediatric scans, low-quality images, or lateral CXRs. […] All outliers were automatically excluded from the dataset […].\"
\nWhen I did some exploratory data analysis, I saw 839 .dicom files with PatientAge given as 000Y (plus 000D = 0 days?), which might either be yet another placeholder for missing or might really mean 0 years of age (i.e. < 1 year-old). I'm thinking that some of these might really be that, since some of the ages are below < 18 years, e.g. there's 107 images with 001Y to 017Y.
I'm curious on whether there's more information in this, because obviously some diagnoses are just much more likely at certain ages.
", "rawMarkdown": "From the paper describing the methods to create the dataset, I see that \"The collected raw data was mostly of adult PA-view CXRs, but also included a significant amount of **outliers such as** images of body parts other than chest (due to mismatched DICOM tags), **pediatric scans**, low-quality images, or lateral CXRs. [...] All outliers were automatically excluded from the dataset [...].\"\n\nWhen I did [some exploratory data analysis](https://www.kaggle.com/bjoernholzhauer/eda-dicom-reading-vinbigdata-chest-x-ray/edit/run/50675891), I saw 839 .dicom files with `PatientAge` given as `000Y` (plus `000D` = 0 days?), which might either be yet another placeholder for missing or might really mean 0 years of age (i.e. < 1 year-old). I'm thinking that some of these might really be that, since some of the ages are below < 18 years, e.g. there's 107 images with `001Y` to `017Y`.\n\nI'm curious on whether there's more information in this, because obviously some diagnoses are just much more likely at certain ages.", "votes": 2 } ], "comments": [ { "id": 1136780, "author_name": "DatNT", "author_url": "", "post_date": "2021-01-03T12:25:02.173000", "content": "I'm one of the papers' authors and yes, this subset is adult-only.
", "votes": 3, "replies": [ { "id": 1136784, "author_name": "Björn", "author_url": "", "post_date": "2021-01-03T12:28:41.193000", "content": "Thanks, so stuff like Patient age ='012Y' in the .dicom meta-data would be a data issue?
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "1136780": "I'm one of the papers' authors and yes, this subset is adult-only.", "1135183": "From the paper describing the methods to create the dataset, I see that \"The collected raw data was mostly of adult PA-view CXRs, but also included a significant amount of **outliers such as** images of body parts other than chest (due to mismatched DICOM tags), **pediatric scans**, low-quality images, or lateral CXRs. [...] All outliers were automatically excluded from the dataset [...].\"\n\nWhen I did [some exploratory data analysis](https://www.kaggle.com/bjoernholzhauer/eda-dicom-reading-vinbigdata-chest-x-ray/edit/run/50675891), I saw 839 .dicom files with `PatientAge` given as `000Y` (plus `000D` = 0 days?), which might either be yet another placeholder for missing or might really mean 0 years of age (i.e. < 1 year-old). I'm thinking that some of these might really be that, since some of the ages are below < 18 years, e.g. there's 107 images with `001Y` to `017Y`.\n\nI'm curious on whether there's more information in this, because obviously some diagnoses are just much more likely at certain ages." } }