Segmentation

The ASME V&V 40 sub-committee is tasked with writing a framework that would guide modelers on how to establish the credibility of their computational models of medical devices.  Estimating the input uncertainty is one of the key steps of the credibility assessment.  One source of error is geometry extracted from medical images.  The CFD-subgroup is organizing a segmentation challenge.  The goal of this study is to estimate the user error associated with the segmentation process.  We are hoping to include users of all the major segmentation codes.

Segmentation Exercise # 2

Background
The first segmentation challenge relied on scan data of a glass aneurysm model, which is used in the development of intravascular neurovascular implants. Results of this challenge were presented at the January 2014 V&V40 meeting in San Antonio. Several main lessons were learned thanks to the contributions of approximately 15 study participants:

  • The resolution at which the scan data was acquired, and any subsequent reconstruction steps prior to the generation of the axial image stack, can have a significant influence on the raw data to be interpreted by the user
  • The less distinct the features of interest (e.g. the lumen of the blood vessel), the greater the amount of judgment required of the user, which leads to variability in the reconstructed geometry
  • Any blurring/smoothing of the image data can reduce the pixel intensity and artificially extend the partial volumes of each pixel, which can be problematic when relying on a pre-determined threshold-based segmentation protocol to segment the data
  • Regions of high curvature are more susceptible to inter-user variability
  • The fluid dynamics consequences between the “truth” and “worst” cases are significant, and illustrate the influence that segmentation, reconstruction, and reference boundary conditions can have on the flow prediction (e.g. velocity driven inlet vs. mass flow rate driven inlet)

After reviewing the results of the first segmentation challenge, the subgroup proposed that the second challenge be based on a clinically acquired CT dataset which better reflects image conditions from which clinically relevant segmentations are constructed. Therefore, the AAA (abdominal aortic aneurysm) patient dataset was selected.

Segmentation Exercise
The second segmentation challenge is focused on the AAA disease state. Aneurysms in this region have catastrophic consequences for the patient when they rupture. The predominant method for treating these aneurysms is to place a stent-graft in order to re-establish minimally thrombotic lumen and structurally reinforce the wall. Imaging is required by physicians in order to visualize the vessel shape and devise a treatment plan.

image 1a   image 1b

 

 

 

 

 

 

 

 

 

Figure 1. (Left) A TAA and AAA is illustrated. (Right) Placement of a AAA stent graft is illustrated. [Wikipedia] 

Tier 1 Participation: Segmentation and Reconstruction of Geometry

1.) The CT image study from the clinical grade CT scanner can be downloaded from the following location: https://www.yousendit.com/download/elNJSU5NR3NJMHM5WThUQw

2.) If you need a license of segmentation software to participate in the exercise, please email Todd Pietila at todd.pietila@materialise.com for a Mimics license or Kerim Genc at k.genc@simpleware.com for a Simpleware license. A temporary license of either software can be arranged for your use. Challenge participants should also feel free to use other 3D image segmentation software, since the goal is to use as many different software packages as possible and characterize variability. Software use will be tracked, but will be kept anonymous with reference to the users.

3.)  Import the CT image study into the image segmentation software of your choice.

image 2

 

 

 

 

 

Figure 2. CT dataset showing the segmentation region of interest (AAA) outlined in green bounding box.

4.) Segment and reconstruct a 3D geometry of the vessel lumen. For the purposes of this exercise, focus on the region of the abdominal aorta and the common iliac arteries (bounded by the rectangular box as illustrated in Figure 2. Please reconstruct the vessel lumen (ie. the volume that blood flows through) as shown in Figure 3. The reconstructed surface should be smooth and free of small branching vessels. Do not include calcium deposits in the vessel wall.

image 3

 

 

 

 

 

 

Figure 3. Reconstructed blood volume of the abdominal aorta and common iliac arteries.

5.) If Simpleware was used for segmentation, please send the STL to k.genc@simpleware.com.  If Materialise or any other software was used, please send the reconstructed geometry in .stl format to Todd at todd.pietila@materialise.com.    Please also mention the segmentation software that was used to reconstruct the geometry and self-identify yourself as a Novice, Medium, or Expert user of segmentation software. We will be compiling statistics of geometric differences between the models.

Tier 2 Participation: CFD Analysis of Flow Through the Reconstructed Geometries

For participants with access to fluid flow simulation software, the next tier of work will involve characterizing intra-aneurysmal flow in the reconstructed geometries. After Todd has collated a number of different geometries from the pool of participants, a second round of emails will be sent out detailing the availability of median, minimum and maximum volume reconstructions.

The intention is for participants to apply their simulation tools and expertise to observe changes in three very basic metrics are of interest: (1) The flow rate of blood entering/exiting the flow domain. (2) A qualitative assessment of intra-aneurysmal flow (3) The distribution of shear stress along the vessel wall (in pictorial format). If you are interested in participating in the second phase of the study, please contact Ricky at Ricky.Chow@lakeregionmedical.com .