ubern-mia / pyradise Goto Github PK

Numpy has deprecated the support for the library-specific np.float type. We should replace it with Python's float.

Because PyRaDiSe will probably be used with common public datasets, we should provide tutorials for converting those datasets to discrete images so that users can reuse the existing code and speed up their work. Maybe we can create a separate tutorial section on public dataset conversion!?

• Harmonize params of SubjectWriter, DicomSeriesSubjectWriter, DirectorySubjectWriter
• Synergize path_checks, dir_checks, etc.
• Enhance test friendliness

I have in a folder both RT Structs and the dose predictions along with the imaging data (CT, MR*) and registration matrices between these. It would be nice to also generate a NIfTI file for the RT Dose (typically named RD*.dcm) from such a structure as well.

The B-Spline Interpolator in the registration process seems to add noise in form of negative values to images with a min value of 0.
Threshold is not an option.

This is in between a hard candy and wish full thinking, since there is a wast amount and variety of sequence types.
It very likely that this request touches other parts of the pipeline as well, presumably some data control/check module (not implemented yet)
Since this is rather a broad topic I try to focus on the registration module for now.
It would beneficial to have a registration protocol, which acknowledges the sequences types.
Especially, this would be helpful to deal with diffusion and perfusion images, where more nuanced registration strategies need to be employed.

iterative execute should return the current filter class name in order to handle interim result.

• Reference segmentation mask to image sequence (Multi masks as well)
• Transform seg mask with ref image transformation matrix
• Use KNN interpolator to preserve seg mask labels

As far as I can tell, intra- and inter-registration do not appear to be linked.
However, it is common for registration to be performed sequentially.
The image sequence closest to a particular atlas sequence is used for an inter-registration step, and then the other sequences are intra-registered with the first sequence already registered.

Therefore, I propose an intra-inter-registration method, to address this issue.

This is a general comment: I would imagine avoiding 'ion' and 'ing' suffixes to package names will make the code read better. This way, the names indicate only what they do, and not the process of doing it (I hope this makes sense).

If you (@ruefene) think I can make this change myself, please let me know here and I can create a new pull request with these changes.

Thanks!

previous subject state.
The playback function is not ideal for this purpose.
The mechanic should cover the following:

• Deep copying a subject and subject tape.
• Adding a subject tag

Feedbacks indicate that PET is used in other institutions and we should add support for it. However, PET images seems to possess a scaling of the intensity values (see DICOM attributes RescaleIntercept, RescaleSlope, and RescaleType) and potentially have additional fields to consider for correct import. Let's dive in this topic with our experts.

@ruefene - the more I think of this, the more it feels like typical users would not really connect with the name: serialization to save data to files. I understand it is technically correct, and along the same lines, you could also name the loading package 'deserialization'.

However, to make this easier to read and understand, I propose combining the two and renaming the entire package 'fileIO' or something similar.

I imagine

from pyradise import fileio

fileio.load(...)
fileio.save(...)

to look cleaner than

from pyradise import serialization 
from pyradise import loading

....