tomelse / patato Goto Github PK

Currently, the field of view definition for the reconstructions made in PATATO and the initial pressure distributions extracted from PATATO are not consistent. How to fix this for regions of interest if there are 3D initial pressures but 2D reconstructions?

Note: this will require changing some functions from np.mean to np.nanmean

Need to get PyPi distribution working. toml/setup.py seem to have some sort of conflict.

The imports have been really slow recently - just annoying not a deal breaker

The implementation is a bit hacky and very specific to the data types we commonly come across. The use of xarray is a bit problematic with e.g. JAX/PyOpenCL/HDF5 arrays. Slicing is also a bit dodgy, although lazily evaluated, which is handy.

The goal is to write a custom unmixer that uses the LSTM as implemented in https://github.com/jgroehl/learned_PA_sO2_estimator to estimate sO2 for an input image.

import patato <- this is really slow.

Read a data file.
Create reconstruction.
Export to IPASC format.
Reload from IPASC format.
Create reconstruction.

Compare both reconstructions for equivalency.

https://setuptools.pypa.io/en/latest/userguide/datafiles.html

A number of issues seem to occur on Windows: slow data loading, JAX installation is complicated.

Run tests to verify that ROIs are being displayed correctly and that masks are being generated and applied correctly.