Info
Background:
Understanding magnetic nanoparticle (MNP) distribution and accumulation in tumors is essential for advancing cancer nanomedicine. Meta-analyses indicate that a median of only ~0.7% of the injected dose reaches solid tumors, independent of MNP size, shape, composition or surface chemistry [1-3]. At the same time, these studies reveal that only a small fraction of reports include quantitative biodistribution measurements at three or more time points, as established imaging methods often lack the quantitative precision and longitudinal studies required to evaluate MNP biodistribution are difficult to implement. Magnetic Particle Imaging (MPI) is an emerging imaging modality that enables non-invasive and quantitative evaluation of nanoparticle distribution in vivo.

Hypothesis/Objective:
Similar to nuclear medicine techniques, MPI typically requires multimodal image registration with anatomical imaging modalities such as Computed Tomography (CT) or Magnetic Resonance Imaging (MRI). However, defining regions of interest directly from MPI images can be difficult when organs are closely spaced, and anatomical segmentation using CT or MRI can also be challenging due to limited soft-tissue contrast. In this work, we evaluated different approaches for the segmentation of internal organs to quantify MPI tracer biodistribution in mice and we validated the use of an artificial intelligence (AI) tool to generate soft-tissue organ segmentations by comparing cohorts of mice that received liver and spleen contrast agent (CA) with one that did not.

Methods:
Live mice were imaged using CT and MPI in 7 time points (1, 3, 6, 9, 12, 24 and 48 h from injection), with one cohort receiving both MPI tracer and the CT CA (cohort CA+, n=3) and another receiving MPI tracer only (cohort CA-, n=3). The MPI tracer consisted of an in-lab synthesized iron-oxide magnetic nanoparticle coated with PEG and the CT CA was ExiTron nano12000 [4]. A pilot study was conducted to determine optimal ExiTron nano12000 dosage (5 CA doses, n=1 for each), which was chosen as a 1:1 dilution of ExiTron nano12000 in saline solution. Brain, heart, liver and spleen segmentations were obtained from MPI scans by thresholding inside a user-selected, organ-specific rectangular region of interest (ROI) and compared with CT-based segmentations created using SlicerNNInteractive [5] within 3D Slicer, an AI framework for interactive 3D segmentation that enables rapid voxel-wise segmentation using 2D prompts.

Results:
The resulting CT-based segmentations were compared with MPI-thresholds of tracer intensities and ex vivo organ shape to evaluate the correspondence between anatomical structures and MPI signal distribution for the different cohorts (see Fig. 1). The use of SlicerNNInteractive to generate CT-based segmentation was validated by comparing segmentation shape from the CA+ and CA- cohorts as well as the ex vivo shape and volume of the organs, showing that the administration of CA is necessary to get anatomically correct segmentation of soft-tissue organs as the liver and spleen but not necessary for organs with natural contrast as the brain (surrounded by the skull) and the heart (surrounded by the air in the lung). Circulation half-life was obtained from both groups from the heart segmentations, where both methods show consistent results. Accumulation and elimination rates were calculated for the liver and spleen, where the segmentation approach drastically affects results.

Conclusion:
Combining CT contrast enhancement with AI-assisted segmentation enables reliable and reproducible delineation of organs necessary for biodistribution and pharmacokinetic studies. These results highlight the value of multimodal segmentation approaches for improving quantitative analysis of nanoparticle biodistribution and supporting the development of particle-based therapies.

Image/Figure:
Click to View

Image/Figure Caption:
Figure 1. (A) CT scans of a mouse in the CA+ cohort at different time point since the CA injection and volume of the liver and spleen normalized to the ex vivo volume for different dilution factors of the CA (pilot). (B) Organ segmentations created with nnInteractive for a representative mouse in the CA- and the CA+ cohort, along with the liver volumes as a function of time for all the mice in both cohorts, showing that CA+ is needed to obtain a constant volume throughout. (C) Organ segmentations obtained with nnInteractive and MPI-based thresholding for a representative mouse in the CA+ cohort, overlayed on a 3D MPI scan. (D) Liver volumes obtained for all the mice in the CA+ cohort with both nnInteractive and MPI-based threshold segmentations, showing nnInteractive is more precise.

Author
Daniela Paola Valdés
University of Florida