Siemens mMR single scatter has unexpected scale

[KrisThielemans]

As observed in #1163, the scatter scale factors for the NEMA phantom had to be around .2, but normally STIR single scatter is scaled to have the correct absolute values. #1279 fixed the thresholds for the scale factor, such that the example now works, but it is still unclear why the scatter scales are so much lower than expected.

[KrisThielemans]

This is still a problem for SIRF as it uses the default settings for the scatter estimator

STIR/src/scatter_buildblock/ScatterEstimation.cxx

Lines 88 to 89 in bdf04ac

	this->min_scale_value = 0.4f;
	this->max_scale_value = 100.f;

[markus-jehl]

@KrisThielemans : can this be closed?

[KrisThielemans]

Sadly not. We still don't know why the single scatter estimate is wrongly scaled for the mMR.

[KrisThielemans]

This was indeed fixed by #1531. For the NEMA IQ data used in the examples, I now get the following scale factors (before filtering) when disabling all thresholding:

Corresponding axial profile through the reconstruction image (averaged over a central square of size 40x40 voxels):

where the bump around 70 is due to the spheres.

Note that there is no bed in the current mu-map.

Plots generated with

import numpy
import matplotlib.pyplot as plt
import os
os.system("sed -n '/upsample_and_fit_scatter_estimate.*before/{;n;p;}' output/scatter.log|tr -d '{}' > scalefactors_before_threshold.csv")

b=numpy.loadtxt('scalefactors_before_threshold.csv',delimiter=',')
plt.figure()
plt.plot(b.transpose());
plt.legend(['1','2','3','4','5'])

import stir
im=stir.FloatVoxelsOnCartesianGrid.read_from_file('final_activity_image_42.hv')
import stirextra
arr=stirextra.to_numpy(im)
p=numpy.mean(arr[:,120:160,120:160],axis=(1,2))
plt.figure()
plt.plot(p)
plt.show()