search

QIICR / lidc2dicom

Scripts for converting TCIA LIDC-IDRI collection derived data into standard DICOM representation from project-specific XML format.
25 stars 12 forks source link

Fix Image Paths #6

Closed CPBridge closed 3 years ago

CPBridge commented 3 years ago

This PR fixes/updates the method of determining the input directory to ensure it works with the current form of the LIDC dataset.

I recently tried to run the script with the latest version of pylidc (0.2.2) installed and a recently-downloaded version of the LIDC-IDRI dataset and had a problem with script using incorrect paths to the input image files.

It appears that the script assumes that the image paths appear at 

<imagesDir>/<subject_id>/<studyUID>/<seriesUID>

but this does not match the directory hierarchy of the dataset downloaded from TCIA. Presumably it did at some point and something changed but I haven't managed to pinpoint the specifics. Currently the downloaded dataset uses what looks like study and series descriptions in the hierarchy rather than UIDs.

In order to fix this I simply set the input image location according to the scan.get_path_to_dicom_files() method of pylidc's Scan method and I was able to run the script without further problems.

fedorov commented 3 years ago

@CPBridge thanks for the PR, looks good! But can you clarify your goal? This repo was used to do the conversion, which was then submitted and archived by TCIA here: https://wiki.cancerimagingarchive.net/display/DOI/Standardized+representation+of+the+TCIA+LIDC-IDRI+annotations+using+DICOM.

It is not a script that is intended to be used again, but rather included for posterity.

CPBridge commented 3 years ago

Hi @fedorov understood, I already have your converted files downloaded. I'm trying to adapt this to use pure python tools (https://github.com/MGHComputationalPathology/highdicom) and as a starting point wanted to reproduce your results using your toolchain and go from there. Thought the fix might be handy just in case anyone wants to do something similar in the future for whatever reason. Thanks for clarifying all the same, and thanks for open-sourcing the script!

fedorov commented 3 years ago

as a starting point wanted to reproduce your results using your toolchain and go from there

Makes sense! I would definitely be interested to see the result of your conversion into a highdicom-based script!

CPBridge commented 3 years ago

I'll be sure to let you know when I have it working! Thanks