MRI and Open-Source Python Tool Reveal How Subcutaneous Drug Depots Really Form

Subcutaneous injection has become one of the most popular non-oral routes for delivering biotherapeutics, largely because patients can often self-administer doses that would otherwise require an intravenous visit to a clinic. Yet despite this growing popularity, what actually happens to the injected solution once it enters the fatty tissue beneath the skin has remained surprisingly poorly characterized. A new open-access study published in AAPS Open on 1 September 2026 addresses that gap with an integrated experimental and computational platform that combines an ex vivo porcine skin model, magnetic resonance imaging, and open-source Python tools to visualize and quantify subcutaneous drug depots in three dimensions.

The research team, led by Shuying Wang and including Mikhail D. Murashov, Michal R. Tomaszewski, Tycho Heimbach, Aparna R. Chakravarti and Corin O. Miller, chose porcine abdominal wall tissue because pig skin closely resembles human skin in structure and thickness, with similar epidermal, dermal and subcutis layers. Fresh tissue samples measuring two inches by two inches by 1.25 inches were placed in the bore of a 7 Tesla small-bore Bruker BioSpec scanner, where injections were performed directly into the tissue. T2-weighted turbo spin echo sequences with fat suppression achieved an in-plane resolution of about 0.234 millimeters with 1-millimeter slice thickness, and no contrast agents were needed. The entire injection-and-scan cycle took roughly five minutes per sample, with imaging beginning immediately after the injection was completed.

The computational side of the workflow is what distinguishes this study from earlier MRI-based depot imaging efforts. Previous work, including studies by Jucker and colleagues on cabotegravir in rats and healthy volunteers and by Pettis and colleagues on large-volume saline injections in healthy adults, demonstrated that MRI can localize and characterize subcutaneous depots with high spatial resolution, but neither provided a standardized method for three-dimensional rendering and quantitative assessment. Many quantitative parameters used in clinical modeling of post-injection solution depots remain theoretical, based largely on empirical assumptions with limited experimental validation. The new method aims to close that gap with a reproducible, freely accessible framework.

The segmentation step relies on a region-growing, or seed-growing, algorithm implemented in the SimpleITK library. A user manually selects a single seed point within the depot on an MRI slice and sets upper and lower grayscale thresholds; the lower threshold is anchored to the background signal so surrounding tissue is excluded, while the upper threshold is typically set beyond the maximum intensity to ensure the full depot is captured. The algorithm then expands automatically from the seed into adjacent pixels in all directions until it encounters pixels outside the thresholds. Once defined, the parameters can be applied across all slices with minimal adjustment, eliminating manual slice-by-slice contouring and reducing the unintentional bias that manual delineation introduces.

After segmentation, the binary slices are stacked into a three-dimensional mesh using the meshlib library, with voxel dimensions of roughly 0.234 by 0.234 by 1 millimeter reflecting the anisotropic imaging resolution. The PyVista library renders the mesh interactively, and a bounding box yields the depot’s dimensions along each axis. A smoothing filter fills fine void spaces within the image to avoid overestimating the external surface area, with the smoothing factor held constant for each injection volume. From the rendered mesh, the team computed the effective depot volume, surface area, and surface-to-volume ratio, the last of which reflects how widely the depot disperses within the tissue.

The study comprised two investigations. Study 1 examined formulation attributes using three semaglutide concentrations, 1, 3 and 10 milligrams per milliliter, prepared to match the buffer composition in the Ozempic package insert. In Study 1.1, concentrations and injection volumes were varied together to hold the total dose constant at 0.5 milligrams, using 50 microliters of the 10 mg/mL solution, 170 microliters of the 3 mg/mL solution and 500 microliters of the 1 mg/mL solution, each run in triplicate. In Study 1.2, a fixed 100-microliter volume was injected at each concentration, with three or four replicates per group. Injections were delivered at a consistent 45-degree angle using a custom 3D-printed guide after the skin was tented with a finger pinch.

A consistent and notable finding emerged across all groups: the segmented effective depot volumes were consistently larger than the injected volumes, by a factor of 1.2 to 1.6. In Study 1.1, the average ratio of injection volume to rendered depot volume was 0.64, 0.62 and 0.81 for the 10, 3 and 1 mg/mL groups respectively. The authors attribute this apparent expansion to the partial volume effect: as liquid disperses within fresh tissue, voxels that are only partially filled are counted as part of the depot, inflating the reconstructed volume. They note that in live animals the measured depot volume would likely be even larger owing to inflammatory tissue response, though some prior MRI studies did not observe a similar effect, possibly due to differences in segmentation methods, imaging parameters or injectate properties.

Depot morphology also differed with formulation concentration. Depots were generally non-spherical, expanding in proportion to injection volume, with shape variability attributed to the heterogeneity of the local tissue topography and thickness; in some extreme cases the depot became irregular when its base penetrated past the fat layer into fiber tissue of the muscle layer or intramuscular space. Surface areas followed the same rank order as injection volumes, while surface-to-volume ratios showed the opposite trend, and the 1 and 10 mg/mL formulations differed significantly in both surface area and surface-to-volume ratio at P less than 0.01. In Study 1.2, with volume held constant, depot volumes were close to or slightly larger than 100 microliters with no significant differences among concentrations, but the lowest concentration showed the greatest variability in segmented volumes, while the 10 mg/mL formulation produced narrower distributions and a more compact, consistent depot shape, a result the authors caution was supported by an insufficient sample size to validate confidently.

Study 2 compared devices, injecting 1 milliliter of semaglutide at the commercial Ozempic concentration of 1.34 mg/mL or Praluent at 150 mg/mL alirocumab through either a prefilled syringe or a Molly autoinjector, with matched 12.7-millimeter 29-gauge needles. Autoinjector injections were delivered at 90 degrees, prefilled syringe injections at 45 degrees with a skin pinch, and each condition was duplicated. Mean depot volumes exceeded injection volumes for both devices and both injectates, and although autoinjector depots were slightly larger on average, the differences were not statistically significant. The autoinjector-generated depot surface appeared less smooth for alirocumab, which the authors suggest could reflect the spring-driven injection force of the device, potentially stronger than manual thumb-finger injection, though they also note that the high-viscosity monoclonal antibody formulation might contribute and that viscosity was not directly measured. With only two replicates per group, the authors explicitly state that the sample size precludes definitive conclusions.

The team is careful to frame the work as a proof of concept. The ex vivo model cannot capture dynamic in vivo processes such as temporal changes in depot morphology, tissue perfusion, or inflammatory responses, all of which may influence depot evolution and drug absorption. The formulation and device variables examined were not exhaustive, and the study covered injectable solutions only, not crystalline or polymeric suspensions, in situ forming implants, or large-volume injections with hyaluronidase. The anisotropic voxel resolution, with coarser spacing in the z-plane, may reduce quantitation accuracy for smaller injection volumes, and potential micro-crack formation in the adipose tissue was not assessed. Within those limits, the platform demonstrated applicability across two therapeutic modalities, a peptide and a monoclonal antibody, spanning injection volumes from 0.05 to 1 milliliter and formulation concentrations from 1 to 150 mg/mL, and by relying on open-source libraries it removes technical and financial barriers for laboratories without access to paid software, offering the subcutaneous drug delivery community a versatile, reproducible tool for formulation and device development.

Subject of Research: MRI-based computational characterization of subcutaneous drug depots in an ex vivo porcine skin model

Article Title: MRI-based computational characterization of subcutaneous drug depots in ex vivo porcine model: implications for biotherapeutics and delivery devices

Article References: Wang, S., Murashov, M. D., Tomaszewski, M. R., Heimbach, T., Chakravarti, A. R., & Miller, C. O. (2026). MRI-based computational characterization of subcutaneous drug depots in ex vivo porcine model: implications for biotherapeutics and delivery devices. AAPS Open, 12(1), Article 56. https://doi.org/10.1186/s41120-026-00191-1

Image Credits: AI Generated

DOI: 10.1186/s41120-026-00191-1

Keywords: subcutaneous injection, drug depots, MRI, semaglutide, alirocumab, porcine skin model, biotherapeutics, autoinjector, prefilled syringe, image segmentation, drug delivery, 3D rendering