We recently demonstrated the value of
reconstructing SPECT data with fully 3D Monte Carlo
reconstruction (F3DMC), in terms of spatial resolution and
quantification. This was shown on a small cubic phantom (64
projections 10 x 10) in some idealistic configurations. The goals of
the present study were to assess the effect of noise and modeling
errors on the reliability of F3DMC, to propose and evaluate
strategies for reducing the noise in the projector, and to
demonstrate the feasibility of F3DMC for a dataset with realistic
dimensions. A small cubic phantom and a realistic Jaszczak
phantom dataset were considered. Projections and projectors for
both phantoms were calculated using the Monte Carlo simulation
code GATE. Projectors with different statistics were considered
and two methods for reducing noise in the projector were
investigated: one based on principal component analysis (PCA)
and the other consisting in setting small probability values to
zero. Energy and spatial shifts in projection sampling with
respect to projector sampling were also introduced to test
F3DMC in realistic conditions. Experiments with the cubic
phantom showed the importance of using simulations with high
statistics for calculating the projector, and the value of filtering
the projector using a PCA approach. F3DMC was shown to be
robust with respect to energy shift and small spatial sampling offset
between the projector and the projections. Images of the
Jaszczak phantom were successfully reconstructed and also
showed promising results in terms of spatial resolution recovery
and quantitative accuracy in small structures. It is concluded that
the promising results of F3DMC hold on realistic data sets.