Simulation-embedded MLEM reconstruction using SIMIND for quantitative myocardial SPECT.
TL;DR
Simulation-embedded reconstruction enabled integrated physical modeling within iterative reconstruction and improved the agreement with ideal reference images under controlled phantom-based conditions and supported the technical feasibility and basic validity of acquisition-consistent Monte Carlo-embedded reconstruction for myocardial SPECT.
Abstract
Objective
To implement and technically validate a simulation-embedded maximum likelihood expectation-maximization (S-MLEM) framework for myocardial single-photon emission computed tomography (SPECT) that uses the simulation of imaging nuclear detectors (SIMIND) Monte Carlo code as a forward projector for integrated modeling of attenuation, scatter, and resolution degradation.
Methods
SIMIND was configured to model a clinical SPECT system and an HL-type myocardial phantom. Agreement between the measured and simulated data was assessed using energy spectra, projection and short-axis images, normalized mean squared error (NMSE), and structural similarity index (SSIM). The validated model was incorporated into S-MLEM. Correction performance was evaluated using simulated four-sphere phantom data and three independent HL-type phantom acquisitions. The uncorrected MLEM and S-MLEM images were compared with ideal references obtained from SIMIND without attenuation, scatter, collimator blurring, or statistical noise. The evaluation included profiles, NMSE, SSIM, and myocardial count uniformity, expressed as the coefficient of variation (CV) of the polar-map values.
Results
SIMIND closely reproduced the measured data. In a four-sphere phantom, S-MLEM improved deep-source count recovery, reduced superficial shape distortion, and restored isotropy. In the HL-type myocardial phantom, S-MLEM improved basal and inferior wall depiction, sharpened the myocardium-cavity interface, reduced cavity spill-in, increased SSIM, and reduced NMSE. Polar map CV decreased from 12.77 ± 0.30% with uncorrected MLEM to 6.48 ± 0.32% with S-MLEM, approaching the ideal reference value of 5.70%.
Conclusion
Simulation-embedded reconstruction enabled integrated physical modeling within iterative reconstruction and improved the agreement with ideal reference images under controlled phantom-based conditions. These findings supported the technical feasibility and basic validity of acquisition-consistent Monte Carlo-embedded reconstruction for myocardial SPECT and provided a basis for further evaluation under more realistic and clinical conditions.