Dear challenge organizers,
I'm currently walking through the cases descriptions available on TCIA (https://wiki.cancerimagingarchive.net/pages/viewpage.action?pageId=64685580#64685580bcab02c187174a288dbcbf95d26179e8) and I have a few questions about the fields in the csv files.
Both, boxes-train.csv and labels-train.csv contain View field. For some samples the value is suffixed with a number (e.g., "lcc1"). Could you please clarify how it is different from the non-suffixed values? Moreover, could you please indicate, whether the View field contain the "corrected" orientation, or is it the same as in DICOM?
The labels-train.csv file has a Subject field containing values 0 and 2. Could you please clarify its meaning?
Hi Mike, I have forwarded your thread to the people more familiar with the dataset than I am. Best of luck with the Challenge!Posted by: kdrukker @ Dec. 20, 2020, 10:31 p.m.
Thank you for participating in the DBTex challenge.
Regarding the views suffixed with a number (e.g. "lcc1"), they come from studies that contain more than one image of the same view.
In some cases, both views are available (e.g. study DBT-S00905), and in other studies (e.g. DBT-S01508) only one.
Potential reasons for rejection are described in the dataset paper: https://arxiv.org/abs/2011.07995
The View field in the shared CSV files is correct.
It is used to read images in the correct orientation that matches the coordinates of bounding boxes: https://github.com/MaciejMazurowski/duke-dbt-data/blob/master/dcmread_image.ipynb
For some cases, view laterality in the DICOM header may be incorrect and you should rely on what is in the CSV files.
The labels-train.csv file should not have a Subject field, but it is included in the boxes-train.csv file.
The Subject column encodes a radiologist who annotated the box.
Let me know if you have further questions,
We have a forum dedicated to this competition and the dataset in general here: https://www.reddit.com/r/DukeDBTData/Posted by: mazurowski @ Dec. 23, 2020, 4:46 p.m.