AngioLab

From user's Wiki!
Jump to: navigation, search

The AngioLab suite, part of the GIMIAS platform, allows the user to perform an entire workflow from medical images to morphological characterization of the aneurysm sac and parent vessels, virtual treatment and advanced visualization of fluid dynamic simulation results. These tools aim to assist the clinician in the assessment of aneurysm rupture risk and choosing the best treatment option for a specific patient.

AngioLab is meant to be used at the decision-making stage of the type of treatment to be undergone by a given patient. AngioLab provides additional information to the clinician by using the patient's medical images, such as aneurysm morphological characterization and simulation of various types of treatment. It includes the possibility to simulate and visualize the virtual treatment with minimally invasive techniques, such as stenting and coil embolization. These last two are called Virtual Stenting and Virtual Coiling, respectively. AngioLab also includes the necessary tools to perform virtual treatment of intracranial aneurysms, such as medical image segmentation, mesh editing tools and building centerlines and morphological characterization of intracranial aneurysms.

The suite is divided in two main workflows:

  • Morphological Analysis
  • Virtual Device Deployment

DICOM image viewing and PACS connection

This plugin can load medical images in DICOM format acquired from radiological examinations, from machines of different modalities, such as computed tomography angiography (CTA), magnetic resonance angiography (MRA), 3D rotational angiography (3DRA). This module is DICOM-compatible, and thus is prepared to work with integrated health information systems such as PACS.

AngioLab allows to set parameters and connect to a PACS (Picture Archive Communication System) and run the query operations, retrieve and send, using the communication protocol and the services provided by the DICOM Message Service Element (DIMSE). Depending on the configuration of the PACS server you want to connect to, others may also be required authentication parameters for each of the services you want to use through AngioLab. AngioLab allows you to search on the following fields: Patient Name, Patient ID, Patient Birth Date, Study ID and date of the study.

The display module allows three-dimensional reconstruction of high quality DICOM images interactively. In this manner allows display of images by multiplanar reformation, in view orthogonal cuts, cuts multiple views and volume rendering. The viewing window allows adjustment of the level of contrast and brightness, the display of images with all demographics, zoom and pan, rotation, selection of areas of interest (Regions of Interest - ROI). Volume rendering can be performed using standard techniques, such as Volume Raycast 3D Texture Mapping or volumetric image, together with a transfer function optimized for angiography images. The accuracy and speed of volume rendering can be adjusted enabling or disabling the acceleration of the Graphical Processing Unit (GPU) and the level of detail (Level of Details - LOD). Volume rendering includes the option of displaying the volumetric image with a 2D Texture Mapping technique which simulates the acquisition technique of X-ray, so as to obtain images similar to those obtained by fluoroscopy (see Figure 2).

Figure 2. Volume rendering angiography imaging techniques using 3D Texture Mapping and GPU-acceleration (left) and 2D Texture Mapping technique which simulates the X-ray acquisition (right)

Image segmentation

Segmentation from a medical image is used to extract information about a structure or anatomy of interest to the clinician, in our case, is the 3D representation of the intracranial vessels.

The technique used in AngioLab consists of an algorithm based on geodesic active regions (GAR) [1] and is optimized for segmentation of vascular tissue through the standardization of image intensity [2]. The GAR is a geometric deformable model that is governed by a power function where the internal force is defined by the local curvature of the surface to extract and combine external regional and gradient descriptors that drive the evolution of the model to the vascular limit. Regional descriptors are derived from a training set, which is a collection of images that have already been through manual segmentation and have a defined depth between what is background and what is blood vessel.

Figure 3. Comparison between manual measurements (left column) and measurements obtained from the segmentation GAR (right column) of a blood vessel (top row) and aneurysm (bottom row). Figure adapted from [2].

This technique uses different training sets, which enable segmentation of blood vessels from medical images from different modalities: 3DRA, CTA, MRA. Finally, the segmentation surface consists of a mesh composed of triangular elements (see Figure 4).

Figure 4. Segmentation obtained through AngioLab's segmentation algorithm.

There are very few occasions when the GAR algorithm fails, but for these occasions an auxiliar segmentation algorithm has been added. This method is called Otsu and it performs a segmentation that depends on the histogram of gray levels of an image [3]. The technique assumes that only two classes of voxels exist (the blood vessel and background) and calculates the optimum threshold separating those two classes. Generally, the segmentation from Otsu method is very crude and contains many flaws, therefore it should only be used when GAR segmentations fails.

Mesh editing tools

Depending on the quality of the image and the morphology of the patient's vessels, the mesh obtained with the automatic segmentation (GAR) may need some manual corrections. This may occur when, for example, there are arteries that fuse with the aneurysm or when the contrast has failed to completely fill the artery to be studied, to name a few (see Figure 5).

Figure 5. Common problems on a segmentation: Fusion of an artery with an aneurysm (left), incomplete filling of contrast within the aneurysm (red arrow), and fusion artery-artery (right).

The tools availabl in AngioLab are:

  • Global Tools (see Figure 6):
    • Ring Cut: for cutting a vessel using a plane, perpendicular to the vessel, from a point selected by the user on the surface of the vessel.
    • Holes Filling: allows to close all the holes on a surface or choose only a few to be closed.
Figure 6. Tools for global mesh editing: Ring cut (left) and Hole filling (right).
  • Local Tools (see Figure 7):
    • Local surface editing: allows the selection of elements on a triangular surface mesh using different selection techniques (brush, unique triangle selection and sphere) for local editing operations (delete triangles, refine and smooth mesh).
Figure 7. Tools for local mesh editing: Surface selection by sphere (left) and brush (right).

After making the necessary corrections and edits, the result is a proper vascular model for the following steps: morphological analysis, building centerlines and virtual treatments.

Tools for building centerlines

This tool allows the creation of central lines from the vascular model surface mesh, as obtained from the previous mesh editing step.

This tool uses a specific algorithm for blood vessels which is based on the creation of central lines using a Voronoi diagram [4]. The Voronoi diagram is used to define center points from the maximum inscribed spheres along a geometry which is, in this case, a blood vessel (see Figure 8). It is said that an inscribed sphere is maximum when it does not let other inscribed spheres to contain it. From these key central points, the shortest paths are defined between source points and target points. These paths are the centerlines.

Figure 8. Vascular model of a carotid artery (left), the Voronoi diagram (middle) and the centerlines obtained from the Voronoi diagram (right). Figure obtained from [4].

The algorithms used are part of VMTK AngioLab (Vascular Modeling Toolkit) [5]. This tool is especially necessary for the virtual treatment of an aneurysm with stent.


Morphological Analysis

To perform a morphological analysis, it is necessary to extract a simplified geometry of the aneurysm and its closest vessels from the vascular model obtained during segmentation. Once this is obtained, the aneurysmal sac is isolated by a manual selection of the aneurysm neck.

From this simplified geometry, different measurements are calculated such as aspect ratio, non-sphericity index, aneurysm volume, aneurysm surface, surface of the aneurysm neck, width of the aneurysm neck (see Figure 9). The aspect ratio is a measurement that correlates the height of the aneurysm to the width of the aneurysm neck, while the non-sphericity factor relates the aneurysm volume with its area.

Figure 9. Morphological descriptors obtained AngioLab aneurysm.

In addition to these measurements, the calculation of descriptors called dimensional invariant Zernike moments [6] is also performed. These descriptors are used to characterize the shape of the aneurysm and its nearby vessels, without depending on the orientation or scale of the geometry. This means that, to some level of detail, you could be able to rebuild the geometry from the descriptors.

The calculation of the Zernike moments is used to provide a comparison between the new patient's aneurysm with other aneurysms that have been treated previously, their geometric descriptors being stored in a database (see Figure 10) . Being able to list similar aneurysms, could offer further support to the clinician when making therapeutic decisions.

Figure 10. An aneurysm being compared with ten other similar in morphology using the calculation of the Zernike moments.

All descriptors are calculated automatically in real time and are easily stored in a database.

Also, you can refine your search depending on the location (near the posterior communicating artery, ophthalmic artery, etc.), and/or type of aneurysm (saccular or fusiform).

Virtual Stenting

The stent deployment within the arterial geometry is performed by a method called 'Fast Virtual Stenting', developed by Larrabide et al [7]. This technique is based on deformable models, where a partial differential equation of second order is used to deform a mesh under the effect of physical constraints such as the geometry, length and radius of the stent.

Figure 11. Deployment of an Enterprise stent (left) and a SILK stent (right) within an idealized geometry of a bifurcation aneurysm. Figure adapted from [7].

The Virtual Stenting technique requires as input the 3D model vessel segment and a centerline from which to deply the stent, with the restriction of the geometry patient's vasculature. The user can select a point on the centerline, marking the center from where the stent would be deployed for treating an aneurysm.

Once the stent is deployed, the user can view the initial and final position of the stent within the vascular geometry. IF the deploymetn wasn't satisfactory, the stent can be reployed to the desired position. The output of this module is a deployed stent mesh, which can be used for visual understanding of a paticular stent peplyment and flow simulations (see Figure 11).

In this tool, the geometries for the following stents are available: Neuroform (Boston Scientific), Enterprise (Cordis), SILK (Balt Extrusion) and PED (ev3).


Virtual Coiling

Virtual Coiling is a dynamic route planning technique that mimics the insertion of coils into a three dimensional model of an aneurysm, giving plausible distribution of coils within a specific patient anatomy [8]. Essentially, the technique involves the following steps:

  • Insertion of coils, by moving the tip of the coil along an already established path, initializing an initial position and direction within the region where the coils are to be placed.
  • Prohibits coil migration out of the aneurysmal area and coil crashes, for proper construction of the path.
  • If the tip of the coil was trapped between other coils, the technique allows the removal of the coil tip to redirect the advance.

As a result, the Virtual Coiling is capable of obtaining a representation similar to the coils inserted in clinical practice that may be used for visual understanding of the treatment and flow simulations (see Figure 12).


Figure 12. Examples of virtual embolization with coils in two aneurysm model obtained from medical images, using the Virtual Coiling technique. Figure adapted from [8].

AngioLab allows the user to visualize the coils within the aneurysm, try different numbers of coils within the aneurysm and know what percentage corresponds to a certain amount of coils.


Clinical report

AngioLab is capable of creating clinical reports with all the information that the user considers appropriate. This is possible either before, or after staring the analysis of the case. The report may include: text, tables (see Figure 13) and screenshots (see Figure 14).

Upon completion of the clinical analysis, the user can review the information that has been added to the report to decide what information is to be eliminated or added. You can also design the final look and print the report in PDF format.

Figure 13. Example of a table containing a morphological measurements of three arteries near the aneurysm.
Figure 14. Example of a screenshot for bifurcation aneurysm untreated (left) and treated with two stents Enterprise and 11 coils. Both vascular geometries are visualized using 2D Texture Mapping.

References

  • [1] Hernandez, M. & Frangi, A. F. Non-parametric geodesic active regions: Method and evaluation for cerebral aneurysms segmentation in 3DRA and CTA. Medical Image Analysis, 2007, 11, 224 – 241.
  • [2] Bogunovic, H.; Pozo, J. M.; Villa-Uriol, M. C.; Majoie, C. B. L. M.; van den Berg, R.; van Andel, H. A. F. G.; Macho, J. M.; Blasco, J.; Roman, L. S. & Frangi, A. F. Automated segmentation of cerebral vasculature with aneurysms in 3DRA and TOF-MRA using geodesic active regions: An evaluation study. Medical Physics, AAPM, 2011, 38, 210-222.
  • [3] Otsu, N. A threshold selection method from gray-level histograms. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9, 62-66.
  • [4] http://www.vmtk.org/Tutorials/Centerlines
  • [5] http://www.vmtk.org
  • [6] Millan, R.; Dempere-Marco, L.; Pozo, J.; Cebral, J. & Frangi, A. Morphological Characterization of Intracranial Aneurysms Using 3-D Moment Invariants. IEEE Transactions on Medical Imaging, 2007, 26, 1270-1282.
  • [7] I. Larrabide, M. Kim, L. Augsburger, M. C. Villa-Uriol, D.Rüfenacht, A. F. Frangi. Fast virtual deployment of self-expandable stents: method and in-vitro validation for intracranial aneurysmal stenting. Medical Image Analysis. 2012. Vol. 16(3):721-30, 2012. DOI: 10.1016/j.media.2010.04.009.
  • [8] H. G. Morales, I. Larrabide, A. J. Geers, L. San Roman, J. Blasco, J. M. Macho and A. F. Frangi, A Virtual Coiling Technique for Image-Based Aneurysm Models by Dynamic Path Planning. Vol. 32(1):119-129. January 2013. DOI: 10.1109/TMI.2012.2219626. IEEE Trans. on Medical Imaging.