Validation of Sinus Drug Delivery Computational Fluid Dynamics (CFD) Modeling with In Vitro Gamma Scintigraphy

Kathryn Kudlaty, William D. Bennett, Landon T. Holbrook, Alyssa Burke, Jihong Wu, Benjamin Langworthy, Jason P. Fine, Charles S. Ebert, Adam J. Kimple, Brian D. Thorp, Adam M. Zanation, Brent A. Senior, Julia S. KimbellarXiv 2020 · DOI · arXiv

Abstract

Background: Chronic rhinosinusitis (CRS) is a prevalent and disruptive disease. Medical management including nasal steroid sprays is the primary treatment modality. Computational fluid dynamics (CFD) has been used to characterize sinonasal airflow and intranasal drug delivery; however, variation in simulation methods indicates a need for large scale CFD model validation.

Methods: Anatomic reconstructions of pre and post-operative CT scans of 3 functional endoscopic sinus surgery patients were created in Mimics™. Fluid analysis and drug particle deposition modeling were conducted using CFD methods with Fluent™ in 18 cases. Models were 3D printed and in vitro studies were performed using Tc99-labeled Nasacort™. Gamma scintigraphy signals and CFD-modeled spray mass were post-processed in a superimposed grid and compared. Statistical analysis using overlap coefficients (OCs) evaluated similarities between computational and experimental distributions and Kendall's tau rank correlation coefficient was employed to test independence.

Results: OCs revealed strong agreement in percent deposition and grid profiles between CFD models and experimental results (mean [range] for sagittal, axial, and coronal grids were 0.69 [0.57], 0.61 [0.49], and 0.78 [0.44], respectively). Kendall's tau values showed strong agreement (average 0.73) between distributions, which were statistically significant (p < 0.05) apart from a single coronal grid in one model and two sagittal grids of another.

Conclusions: CFD modeling demonstrates statistical agreement with in vitro experimental results. This validation study is one of the largest of its kind and supports the applicability of CFD in accurately modeling nasal spray drug delivery and using computational methods to investigate means of improving clinical drug delivery.

Introduction

Chronic rhinosinusitis (CRS) is a persistent inflammatory condition affecting approximately 12% of the United States population. While nasal steroid sprays are considered the "gold standard" for medical management, they fail to provide relief in nearly 50% of patients. A primary factor in this failure is poor drug distribution; the complex, narrow passages of the sinuses are difficult to reach, and medication often deposits prematurely in the anterior nasal cavity.

To address this, researchers have turned to Computational Fluid Dynamics (CFD) to simulate airflow and drug deposition. These "digital twins" allow for the rapid testing of different spray angles, nozzle designs, and surgical outcomes. However, the utility of CFD depends entirely on its accuracy. This study provides a large-scale validation of CFD models by comparing digital simulations against physical experiments using 3D-printed sinus models and radioactive tracers.

Methods and Materials

The study utilized a comparative "twin" methodology across 18 distinct cases derived from the CT scans of three patients at the University of North Carolina at Chapel Hill, covering both pre- and post-operative functional endoscopic sinus surgery (FESS) anatomies.

Digital Simulation

Anatomic reconstructions were created using Mimics™. The researchers employed Ansys Fluent™ to simulate laminar airflow. Drug particles were modeled using a Discrete Phase Model (DPM), which treats the nasal spray as a collection of millions of individual droplets with specific mass and velocity characteristics.

Physical Experimentation

The same patient anatomies were 3D-printed to create high-fidelity physical models. To track deposition, the researchers used Nasacort™ labeled with a Technetium-99m radioactive tracer. After the spray was administered to the 3D-printed models, gamma scintigraphy (radioactive imaging) was used to map exactly where the medication landed.

Statistical Comparison

To compare the digital and physical results, a 3D grid was superimposed over both models. The study employed two primary metrics:

  1. Overlap Coefficients (OC): To measure the similarity between the computational and experimental mass distributions.
  2. Kendall’s Tau Rank Correlation: To test the statistical independence and agreement between the two datasets.

Key Results

The results demonstrated a high degree of correlation between the digital simulations and the physical experiments:

  • Statistical Agreement: The mean Kendall’s tau value was 0.73, indicating a strong and statistically significant agreement (p < 0.05) in almost all cases.
  • Spatial Overlap: The Overlap Coefficients showed the strongest agreement in the coronal planes (mean 0.78), followed by the sagittal (0.69) and axial (0.61) planes.
  • Deposition Patterns: Both models identified the internal nasal valve as a significant bottleneck. While the physical experiments showed slightly higher penetration into the posterior regions (34.74% vs. 27.99% in CFD), the overall distribution profiles were remarkably consistent.

Discussion and Insights

One of the most significant findings was the consistency of the drug distribution "profile." Even when specific grid squares showed minor variances, the overall shape of the deposition was captured accurately by the CFD. This suggests that CFD is a robust tool for predicting how changes in spray orientation or anatomy will shift drug delivery patterns.

The study also highlighted the sensitivity of validation to grid alignment. Shifting the digital grid by as little as 2.38mm (one pixel) could significantly alter correlation coefficients, suggesting that the resolution of experimental imaging (gamma scintigraphy) is a limiting factor in validation precision.

Limitations and Future Directions

While successful, the study noted several areas for refinement:

  • Environmental Factors: The current CFD models did not account for humidity, which can cause droplets to change size in a real human nose.
  • Mechanical Effects: In physical experiments, the spray nozzle may physically stent (open) the nasal valve, an effect not yet captured in the digital simulations.
  • Spray Dynamics: The model treated the spray as a point source; future work could incorporate more dynamic plume characteristics.

Conclusion

This study represents one of the largest validation efforts for sinonasal CFD to date. The strong statistical agreement between digital simulations and in vitro results confirms that CFD is a reliable methodology for modeling nasal drug delivery. This validation paves the way for using computational tools to optimize clinical treatments and design the next generation of nasal delivery devices.

Summary updated

Abstract

Background: Chronic rhinosinusitis (CRS) is a prevalent and disruptive disease. Medical management including nasal steroid sprays is the primary treatment modality. Computational fluid dynamics (CFD) has been used to characterize sinonasal airflow and intranasal drug delivery; however, variation in simulation methods indicates a need for large scale CFD model validation. Methods: Anatomic reconstructions of pre and post-operative CT scans of 3 functional endoscopic sinus surgery patients were created in Mimics(TM). Fluid analysis and drug particle deposition modeling were conducted using CFD methods with Fluent(TM) in 18 cases. Models were 3D printed and in vitro studies were performed using Tc99-labeled Nasacort(TM). Gamma scintigraphy signals and CFD-modeled spray mass were post-processed in a superimposed grid and compared. Statistical analysis using overlap coefficients (OCs) evaluated similarities between computational and experimental distributions and Kendall's tau rank correlation coefficient was employed to test independence. Results: OCs revealed strong agreement in percent deposition and grid profiles between CFD models and experimental results (mean [range] for sagittal, axial, and coronal grids were 0.69 [0.57], 0.61 [0.49], and 0.78 [0.44], respectively). Kendall's tau values showed strong agreement (average 0.73) between distributions, which were statistically significant (p &lt; 0.05) apart from a single coronal grid in one model and two sagittal grids of another. Conclusions: CFD modeling demonstrates statistical agreement with in vitro experimental results. This validation study is one of the largest of its kind and supports the applicability of CFD in accurately modeling nasal spray drug delivery and using computational methods to investigate means of improving clinical drug delivery.
BibTeX
@misc{kudlaty2020validation,
  author = {Kathryn Kudlaty and William D. Bennett and Landon T. Holbrook and Alyssa Burke and Jihong Wu and Benjamin Langworthy and Jason P. Fine and Charles S. Ebert and Adam J. Kimple and Brian D. Thorp and Adam M. Zanation and Brent A. Senior and Julia S. Kimbell},
  title = {{Validation of Sinus Drug Delivery Computational Fluid Dynamics (CFD) Modeling with In Vitro Gamma Scintigraphy}},
  year = {2020},
  eprint = {2007.12071},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2007.12071},
}
Figure