Error in scatter estimation when using SSRB for attenuation factors

[KrisThielemans]

Scatter estimation happens in 2D, with using SSRB for ProjDataInfoCylindrical to reduce noise (for BlocksOnCylindrical, we just use seg 0).

STIR/src/scatter_buildblock/ScatterEstimation.cxx

Lines 165 to 179 in 5da26d8

	ScatterEstimation::make_2D_projdata_sptr(const shared_ptr<ProjData> in_3d_sptr)
	{
	shared_ptr<ProjData> out_2d_sptr;
	if (in_3d_sptr->get_proj_data_info_sptr()->get_scanner_sptr()->get_scanner_geometry() == "Cylindrical")
	{
	shared_ptr<ProjDataInfo> out_info_2d_sptr(
	SSRB(*in_3d_sptr->get_proj_data_info_sptr(), in_3d_sptr->get_num_segments(), 1, false));
	out_2d_sptr.reset(new ProjDataInMemory(in_3d_sptr->get_exam_info_sptr(), out_info_2d_sptr));
	SSRB(*out_2d_sptr, *in_3d_sptr, false);
	}
	else
	{
	shared_ptr<ProjDataInfo> out_info_2d_sptr(in_3d_sptr->get_proj_data_info_sptr()->create_shared_clone());
	out_info_2d_sptr->reduce_segment_range(0, 0);

For prompts, randoms and efficiencies, we use SSRB in its "adding" mode (do_normalisation=false). However, this is incorrect for the attenuation correction factors. (2D ACFs should correspond to exp(lineintegral(mu)), and hence should be very close to seg 0 of 3D ACFs, i.e we should use averaging, not adding)

STIR/src/scatter_buildblock/ScatterEstimation.cxx

Line 661 in 5da26d8

atten_projdata_2d_sptr = make_2D_projdata_sptr(tmp_atten_projdata_sptr, out_filename);

git blame confirms that this bug was introduced in 852c10d when cleaning-up the SSRB lines to replace them with make_2D_projdata_sptr (in #1006, merged 27 Jun 2022, rel 5.1.0). Before this PR, all SSRBs were called with false (i.e. add), except for the 2D ACFs, where it uses true (i.e. average).