Magnetic particle imaging resolution needed for magnetic hyperthermia treatment planning: a sensitivity analysis.

Purpose: Magnetic particle imaging (MPI) is a nascent tracer imaging modality that generates images from magnetic iron oxide nanoparticles (MIONs) in tissue. MPI resolution is a critical input parameter for defining the reliability of simulations-based temperature predictions for magnetic nanoparticle hyperthermia (MNPH). The objective of this study was to ascertain how spatial resolution provided by MPI data affects the reliability of predicted temperatures and thermal dose in simulations using MPI data as inputs.

Methods: Computed tomography (CT) and MPI scans obtained from a tumor injected with MIONs were co-registered to align their coordinates. Co-registered data were used to obtain geometry and volumetric heat sources for computational simulations of MNPH in phantom tumors. In addition to using the MPI-derived in vivo MION distribution (D1) we analyzed two mathematical MION distributions: uniform (D2) and Gaussian (D3). All distributions were discretized into cubic voxels and the data were imported into a commercial finite element bioheat transfer (FEBHT) software for thermal simulations. FEBHT simulations were conducted using the Pennes' bioheat equation using four different MION specific loss power (SLP) values in the range 300-600 [W/g Fe]. The impact on predicted temperature resolution and thermal dose of spatial resolution were assessed by varying the linear voxel density (LVD) from 0.36 to 4.06 [voxel/mm]. Results were compared against the simulation with the highest LVD [4.06(voxel/mm)], where deviations in temperature of ≤ ±1 [°C] and thermal dose coverage ≤ ±5 [%] were deemed acceptable.

Results: The D3 distribution resulted in the highest predicted temperatures, followed by D1 and D2; however, in terms of thermal dose, D1 showed lowest tumor coverage, requiring higher heat output from MIONs than was required for the other distributions studied. The results of the sensitivity analysis revealed that the predicted tumor temperature increased with LVD across all tested SLP values. Additionally, we observed that the minimum acceptable LVD increased with SLP.

Conclusion: Current (preclinical small animal) MPI scanners provide sufficient spatial resolution to predict temperature to within ±1 [°C], and thermal dose coverage to within ±5 [%] for MION formulations having heat output SLP = <370 [W/g Fe]. Higher spatial resolution is needed to achieve a similar precision when MION SLP exceeds 370 [W/g Fe]. We also conclude from the results that assuming a uniform MION distribution in tissue, which has been a common practice in MNPH simulations, overestimates the SLP needed to deposit meaningful thermal dose.