3D Rendering Methodology

Image Credit: Dr. Matthew Bramlet; Jump Trading Simulation and Education Center.

Heart Library

Sponsored by Jump Trading Simulation & Education Center

3D Rendering Quality and Methods

A few notes regarding the methods that follow:

  1. “Image” will refer to the entire 3D data.
  2. Source image data sets must meet minimum image quality criteria as defined in “Imaging guidelines” and be free of artifact that interferes with creation of model.
  3. The mask overlay refers to the process within the rendering software of segmenting the desired portions of the image data set that will become part of the final digital model.
  4. These processes are independent of the desired viewing or printing method and focus only on the process of evaluating the anatomy on the source images and creating the mask overlay that will determine the digital model as interpreted by an expert in the field of congenital cardiology.

Quality Review

To be accepted for review, the DICOM data set must meet the imaging guidelines and the submission form must be filled out in full. The digital model (.stl file or other acceptable format) should fit into one of the three methodology types and accurately represent source DICOM data. Poorly rendered models that deviate too far from the source image will be rejected.

Imaging guidelines:

  1. 3D Imaging datasets with resolution between 1-1.5 mm iso voxels is ideal. Datasets with lower resolution will not be accepted due to the increased risk of false data transmission to the digital model.
  2. 3D imaging dataset must be free of artifact that interferes with components of final digital model
    • artifact within the vessel lumen that does not obscure the vessel wall would be acceptable
    • blooming artifact on an MRI from a metallic implant that alters the vessel or myocardial wall border would not be acceptable
    • flow artifact that renders the identification of vessel wall impossible would not be acceptable
  3. All DICOM or imaging data MUST BE anonymized.

Methods

  1. Solid blood pool segmentation
    • An image where contrast fills the intracardiac and intravascular structures is overlaid with a mask while all other structures are not. This results in a negative cast of the myocardium and vascular structures.
    • This method is typically utilized for extra-cardiac vascular assessment.
  2. Blood pool/myocardial border segmentation.
    • The method starts with the product of method 1 and then applies a universal cast (or mask overlay in 3 dimensions) of a predetermined thickness* to the entire model. Following this step the previous solid blood pool is deleted and you are left with a cast of the method 1.
      *the thickness used in millimeters should be stated.
    • The final step is to open up the ends of the vasculature where the cast has “capped” the vessels.
    • This method is the most commonly utilized method for rendering hearts currently and is good for both extracardiac and intracardiac evaluation.
    • Final verification of segmentation accuracy should directly compare the mask data (3D contour overlay) that will ultimately generate the .stl or similar file as compared to the source image.
  3. Myocardium and vessel wall segmentation.
    • This method can be achieved one of many ways, but the goal is to highlight the myocardium, vascular walls, and intracardiac structures as visualized on the source image.
    • It is important to use standardized methods of thresholding rather than manual placement of mask overlay, specifically at the endocardial-blood pool border. “Drawing” the mask overlay will decrease reproducibility and increase error and should be avoided.
    • This method is currently the most time consuming method, but is designed to result in an accurate replication of the heart as defined by the source image.
    • Since, no minimal wall thickness is defined, there may be structures which become difficult to print and a separate assessment of the appropriate printing process needs to be considered.
    • Final verification of segmentation accuracy should directly compare the mask data, [(3D contour overlay) that will ultimately generate the .stl or similar file] to the source image.