Innovative Patient-Speciﬁc Intranasal Flow Simulations
Take a deep breath. Now, close your mouth and breathe out through your nose. What you have just done, namely expiration through your nasal cavity, is what is simulated in the following study. Every step of the process, from the segmentation of the flow region, to the surface shape reconstruction and more, is illustrated by Dr. Krause. The challenges faced during preprocessing, which arise from shapes so complex that modern CT scanners cannot capture them exactly, are met using Materialise’s Mimics Innovation Suite, in combination with an innovative automated grid generation and initialisation concept. The approach is realized in the open source library OpenLB: a 2D and 3D fluid flow solver based on Lattice Boltzmann Methods (LBM).
Turning CT data into a surface mesh with Mimics and 3-matic
The construction of a surface mesh, based on CT data of the inner nasal cavity, is done in two main steps.
- First, the region of interest is segmented and represented as an STL surface mesh.
- Then, the unresolved fine structures, which are crucial for the flow characteristics, are reconstructed.
For the first step, Dr. Krause applied seeded and dynamic region growing segmentation schemes. By doing so, he was able to identify the region of interest according to a defined color level or a special change. Mimics proved ideal for accomplishing this step and to a large extent, it automated the process.
The second step, the reconstruction of unresolved structures like small air cavities, paranasal sinuses, and thin passages and tissues, was done with the help of physicians, 3D visualization techniques, and 3-matic.
For this study, the small tissue which separates two passages could not be reconstructed completely. However, some captured parts of the tissue formed a needle structure in the segmented surface representation, so the problem could be solved using tools in 3-matic.
Preparing the simulation using a volume mesh
In order to simulate the airflow in the nasal cavity numerically, the volume needed to be parted into small cubes of equal size, also known as voxels. Their vertices, also called nodes, with their linking edges provide the volume mesh needed for the LBM computation. In order to generate this type of volume mesh from a STL surface mesh, Dr. Krause chose the open source Common Versatile Multi-purpose Library for C++, the code of which is included in the OpenLB library.
Before the airflow could be simulated on a computer, Dr. Krause needed to specify the inflow and outflow regions, which are located at the trachea and the two nostrils. They were distinguished from the flow region and the wall of the nasal cavity by assigning certain different numbers (material numbers) to the nodes of the volume mesh. Depending on their distribution, it becomes possible to initialize the simulation algorithm. Yet, the initialization process for LBM algorithms and its automatization poses a challenge since the algorithm set-up for each node depends on the material numbers of all neighboring nodes. Furthermore, for particular material number distributions the initialization is not possible. These difficulties were overcome by a generic concept proposed and realized in the OpenLB project. This concept relies on sophisticated routines which allow the original voxel mesh to be modified where needed without changing its characteristics.
Successfully simulating expiration and diagnosing a disorder
Expirations through the nasal cavity of a European male adult were successfully simulated with OpenLB. The results were validated by comparison with other experimentally and numerically gained results as well as with measurements taken from the patients involved. The visualized simulation results offer new insights into the flow characteristics of human respiration and for one of the research subjects in particular, the patient-specific approach helped to locate a stenosis which is now assumed to be the cause of a severe peripheral obstructive ventilation disorder that had been diagnosed before the study.
We appreciated Mimics’ user-friendly interface and the time it saved us. The graphical user interfaces and automated routines enabled a high-quality segmentation and reconstruction of the complete human nasal cavity. Furthermore, we found the segmentation algorithms to be both robust and accurate. Finally, 3-matic eased the process of reconstructing thin tissue that was only partly captured by the CT scanner.