Image Postprocessing and Volume Rendering

Abstract

Nearly all images produced in a medical imaging department are processed to some degree. Some processing is done to make digital images look more like their film predecessors. Some processing is done to accentuate certain features (e.g., bone and soft tissue kernels in CT) or to provide higher resolution (e.g., MRA). This postprocessing is typically focused on producing images that are visually pleasing. However, postprocessing can also be used to improve the performance of a CAD algorithm, to produce renderings of components of the image set that are more useful. This chapter will begin with very basic image processing functions, and proceed on to advanced techniques that are increasingly being applied as a part of postprocessing enhancements.

Self-Assessment Questions

• 1. Which of the following is an advanced de-noising algorithm?

  • a. Averaging filter

  • b. Nonlocal means

  • c. Gaussian filter

  • d. Median filter

• 2. Morphological filtering is commonly used for?

  • a. registration

  • b. de-noising

  • c. “cleaning up” segmentation

  • d. image display

• 3. Which of the following filters may be applied in real time (before all images are acquired)?

  • a. averaging

  • b. Nonlocal means

  • c. Gaussian

  • d. Anisotropic diffusion

• 4. In a registration algorithm, the metric is used to

  • a. filter the images

  • b. find the global minimum

  • c. segment the images

  • d. determine if the images are in alignment

• 5. Which registration transform is suitable for neuro-imaging registration?

  • a. Thin-plate spline

  • b. Affine transformation

  • c. Rigid-body transformation

• 6. Which of the following is NOT a direct volume rendering technique?

  • a. MIP

  • b. Marching Cubes

  • c. Volume ray casting

  • d. Splatting

  • e. Shear-Warp

• 7. To simultaneously render more than one object in a volume, which of the following volume rendering technique will lead to the best rendering quality?

  • a. MIP

  • b. SSD

  • c. Volume ray casting

  • d. Splatting

  • e. Shear-Warp

• 8. Describe situations in which MIP is preferable to MinIP.

• 9. Describe situations in which MinIP is preferable to MIP.

Pearls 8.24

• Image filtering reduces noise in images, while preserving edge information. Image filtering may be used as a preprocessing step or in real time during imaging.

• Segmentation is the process of identifying regions of interest within an image and is often used for quantification of size and shape of structures. Morphological operators are often used to “clean up” segmentations, filling in holes, and breaking small bridges between structures.

• Registration algorithms are composed of transformations, similarity metrics, and optimization of algorithms. Registration is useful for comparing studies through time and is easily applied in their own imaging and more difficult and body imaging.

• Volume rendering is a visualization technique to display a 2D image of a full 3D data set.

• MIP and MinIP are the simplest and fastest volume rendering method.

• Direct volume rendering can give users details of the object(s) inside a volume, whereas surface rendering shows only the exterior of objects.