Conventional versus Monte Carlo SPECT reconstruction of Lu-177: Toward reduced bias and variance in quantitative...

Monte Carlo (MC)-based single photon emission computed tomography (SPECT) reconstruction utilizes advanced system models that stochastically sample all possible photon interactions within the patient and detector, potentially increasing quantitative accuracy and precision. Despite this, there have been few studies that have rigorously compared conventional SPECT reconstruction and MC-based reconstruction for metrics that are pertinent to radiopharmaceutical dosimetry.

This paper aims to compare conventional reconstruction with hybrid-MC reconstruction and explores accuracy and precision in total activity estimation within various regions of interest in ${}^{177}\mathrm{Lu}$ imaging.

The MC engine Simulation of Medical Imaging Nuclear Detectors (SIMIND) was integrated with the reconstruction software PyTomography to enable MC-based reconstruction. The following are explored: (i) elimination of triple energy window (TEW)-induced bias obtained by using an MC system model and (ii) improvements to precision and effective reductions in required scan time that are attained when using MC-based reconstruction. The study explores multiple acquisitions of simulated and real phantom/patient data.

Conventional reconstruction with TEW induces a positive bias (e.g., 116.3% recovery coefficient, or RC, in a 72 mm sphere from the MC-simulated data) that is not present with MC-based reconstruction (e.g., 101.3 RC%). This source of positive bias raises RC in smaller spheres, effectively canceling with negative bias incurred from finite resolution; in real phantom and patient data, RCs are lower with MC-based reconstruction than conventional, suggesting that MC-based reconstruction is able to remove the additional source of bias caused by TEW. Furthermore, MC-based reconstruction is able to reduce variability between subsequent acquisitions of the same phantom (e.g., reducing variability by $(4.19\pm 1.39)\%$ in 10 mm sphere), thus resulting in reconstructed images comparable to those obtained with longer scan times (e.g., by a factor of $2.78\pm 0.33$ in the 32 mm sphere).

Compared to conventional reconstruction that uses TEW to correct for scatter, MC-based reconstruction for ${}^{177}\mathrm{Lu}$ (i) reduces sources of systematic bias caused by inadequate TEW scatter estimation and (ii) reduced required scan times by reducing total uptake variability in regions of interest across different scans. For this reason, MC-based reconstruction should be preferred to conventional reconstruction in clinical practice.