Question: What image file do you use to import Image Data for RTSTRUCT extraction?

[jmarsil]

Just to clarify, you are getting your original patient info from the original structure set?

[jasbach]

Ref: New_Interface_React branch, backend/extract_DICOM_data.py

Lines 58-59 instantiate empty dictionaries, keyed by patient ID, values are sub-dictionaries.

Patient IDs are matched as the function walks through the directory to ensure files are only paired same-patient.

Function uses the SliceLocation value in the metadata for each image to associate that image array with the position along the Z axis (line 88)

For the structure set, the contour data is stored as triplets, where the third value is the position in the Z axis. That value is stripped out of the data (line 159) and used to map that set of points to a Z axis position.

The function build_parallel_arrays at line 183 takes these dictionaries and uses the Z position of image/contour pairs to generate mask arrays that are aligned with the corresponding images.

Note that if the patient orientation is not orthogonal (such that a contour might have different Z-axis values across the points) this won't work. This comes up sometimes in MRI data but I haven't encountered it for CT (all of my training data was CT).