Automatic Diagnosis of 3D MRA

Development of a System for the Automatic Diagnosis of 3D MRA image data sets

Senter, Ministry of Economic Affairs
Patrick de Koning

Background

Contrast enhanced (CE) Magnetic Resonance Angiography (MRA) provides 3D images of the morphology of the blood vessels in the body. Interpretation of these 3D images is carried out using 2D projection images called Maximum Intensity Projections (MIP). Using this projection images results in the loss of information, because all information on the 3D space is compressed into a 2D image. This results in the over projection of vessels. Without advanced analytical software, the value of MRA is limited and this advanced technology cannot be used to its full potential.

Goals

The goal of the project is find answers to the following questions:

  • Is it possible to segment a 3D CE MRA dataset with minimal user interaction?
  • Is it possible to use a coarse segmentation of a vascular structure in a 3D MRA dataset to improve the visualization of the vessels?
  • Is it possible to use a coarse segmentation of a vascular structure to visualize specific vessel segments in an optimal view?
  • Is it possible to determine the boundaries of the vessel wall accurately in a 3D CE MRA dataset?
  • With what accuracy can we quantify a stenosis?
  • For which anatomical regions can these new algorithms be used?
  • How can the image processing algorithms that will be developed, be applied in a clinically useful software package?
  • Does the newly developed software improve the interpretation of 3D CE MRA compared to the conventional techniques?

Approach

The problem can be broken down to three separate parts. The first part consists of finding the path of a vessel. The second part segments the vessel and the third and final part calculates clinically relevant parameters of the segmented vessel.

1) The first stage can be divided into a user part and an automatic part. The user has to define the beginning and the end of a vessel segment by placing a start- and endpoint.
The placement of the points is done using the orthogonal projections of the 3D data, either using MIP or Closest Vessel Projection (CVP). In order to place a point in 3D space, it has to be placed correctly on at least two projections. By using a z-buffer created during the generation of the projections, we can position the point in 3D using only one projection.

Placing the start and endpoint needed for the detection of the centerline

The automated part calculates a path from start to endpoint. This is accomplished by using a wave front propagation followed by a backtrack and finally a correction to form the centerline of the vessel.
The WaveProp algorithm simulates the propagation of a wave through a medium. The speed of the propagation is locally defined by a speed function, which relates local intensity values to a speed. The movie below shows the propagation of a wave through a 2D image.
The movie below shows the expanding wavefront. The arrival time of the wave at any given point is indicated by the height (z-value). This arrival time is used for the next stage of the centerline detection. The movie show the expansion in a 2D image, the arrival time is the height. The image below shows the input image of the WaveProp with the startpoint indicated.

The result of the WaveProp algorithm on a 2D image. The height of the image represents the arrival time of the wave at each point. You can clearly sea that the wave expands first through the vessel and then through the remaining parts of the image

The backtracking is the second stage of the centerline detection. This algorithm actually finds a path from end- to start-point (that's why it's called backtrack). The detected path is the fastest route from start- to endpoint for the given speed function.
The movie below shows how this backtracking looks like in 2D. The algorithm simply performs a steepest descent. The arrival time image has no local minima, so the backtrack always ends in the point from which the propagation started.

Performing a backtrack on the results of the WaveProp. You can see the red ball rolling down to the starting point.

The third and final stage of the centerline detection is a correction of the backtrack. Because the backtrack is the fastest route from end- to start point, it tends to cut the corners.
It is impossible to determine the actual centerline without a segmentation that tells you what the boundary of the vessel is, so we can only estimate the boundary and move the pathline to the center of this estimate. The pictures show the correction from pathline (purple) to centerline (blue).

2) The second stage of our approach uses a model of a vessel to segment the vessel.
In order to segment the vessel wall, we first perform a Curved Multi-Planar Reformat (CMPR). This transformation turns the curved vessel into a straight vessel in which the detected centerline forms the exact center of the data set.
After the CMPR, we use a model of a vessel (a tube) to fit it to the data. After this segmentation, we can view the result in the CMPR space, or convert it back to our original image.

Curved Multi-Planar Reformat (CMPR) of a carotid artery. This image was generated using a pathline through the common and internal carotid artery.

Fitting of a vessel model onto the CMPR image

Result of the fitting procedure. You can clearly see that the fit is very well to the image.

The result of the fitting process converted from CMPR space to the regular space.

3) The final stage of our 3-stage analysis calculates the clinically relevant parameters that the user can use to make a diagnosis for the patient. A sample of the parameters we can calculate is shown below, where we calculated the cross sectional area (CSA). We can also calculate circumference, mean- min- max- diameter and we can perform a principal component analysis (PCA) to derive the main axes of the contour.

 

Status

Finished

Publications

  1. Janssen JP, Koning G, Koning PJH de, Tuinenburg JC, Reiber JHC. A new approach for the detection of pathlines in X-Ray angiograms: The wavefront propagation algorithm. Int J Card Imaging 18, 2002: 317-324.
  2. Schaap JA, de Koning PJH, Janssen JP, Westenberg JJM, van der Geest RJ, Reiber JHC. 3D quantification visualization of vascular structures in magnetic resonance images. In: volume 1230 of International Congress Series, pages 974-980. Elsevier Science, 2001.
  3. R. Guzman, H. Oswald, A. Barth, P.J.H.de Koning, L. Remonda, K.O. Lovblad, and G. Schroth. Clinical validation of quantitative carotid MRA. In: volume 1230 of International Congress Series, pages 934–937. Elsevier Science, 2001.
  4. Guzman R, Barth A, Remonda L, de Koning PJH, van der Geest RJ, Lovblad KO, Oswald H, Schroth G. Correlation of quantitative mr-angiography of the carotid artery with in vivo measurement during carotid enarterectomy. Proceedings of CARS 2002 Computer Assited Radiology and Surgery. Hrsg. Lemke HU, et al. Springer Verlag. Berlin, Heidelberg. Juni 2002; 917–922.
  5. Janssen J, Koning G, de Koning PJH, Tuinenburg JC, Reiber JHC. A novel approach for the detection of pathlines in X-ray angiograms; the wavefront propagation algorithm. Proceedings of CARS 2002 Computer Assisted Radiology and Surgery. Hrsg. Lemke HU, et al. Springer Verlag. Berlin, Heidelberg. Juni 2002; 808-813.
  6. De Koning, PJH, van der Geest RJ, Reiber, JHC, "1st International Workshop on Coronary MR and CT Angiography. Lyon , France , 1-3 October 2000. Abstracts," Int.J.Card Imaging 2000; 16(3):185-224
  7. de Vries M, de Koning PJ, de Haan MW, Kessels AG, Nelemans PJ, Nijenhuis RJ, Planken RN, Vasbinder GBC, van Engelshoven JMA, van der Geest RJ, Leiner T. Accuracy of Semiautomated Analysis of 3D Contrast-Enhanced Magnetic Resonance Angiography for detection and quantification of aortoiliac stenoses. Invest Radiol 2005;40:495-503.
  8. Schoonman GG, Bakker D, Schmitz N, van der Geest RJ, van der Grond J, Ferrari MD, van Buchem MA. Magnetic resonance angiography of the human middle meningeal artery: implications for migraine. J Magn Reson Imaging. 2006;24(4):918-921.

Abstracts

  1. [Symposium Neuroradiologicum (World Congress of Neuroradiology) 2002]
    Guzman R, Barth A, Reinert M, Oswald H, Lövblad KO, de Koning P, van der Geest R, Remondal L. (2002) Surgical quality of carotid endarterectomy assessed with postoperative MR-Angiography. XVIIth Symposium Neuroradiologicum, Paris , France . Journal of Neuroradiology 29, Suppl. 1, S71
  2. [Societe de neurochirurgie de langue francaise 2002]
    Barth A, Guzman R, Oswald H, Koning P de, Geest R van der, Lövblad KO, Remonda L. Quantification par angiographie-IRM de l’élargissement de la bifurcation carotidienne obtenue par suture continue sans patch après thrombendartérectomie.
  3. [AANS ANNUAL MEETING 2001]
    Barth A, Guzman R, Oswald H, Koning P de, Geest R van der, Lövblad KO, Remonda L. Magnetic Resonance Angiographic measurement of vessel lumen expansion after carotid endarterectomy closed with a running suture without patch.
  4. [AANS ANNUAL MEETING 2002]
    Barth A, Guzman R, Oswald H, Lövblad KO, Koning P de, Geest R van der, Remonda L. Quality control of carotid endarterectomy with postoperative MR-angiography.

Contact

Patrick J.H. de Koning, MSc.
Division of Image Processing
Department of Radiology, 1-C2S
Leiden University Medical Center
P.O. Box 9600
2300 RC Leiden
The Netherlands
Tel. +31 (0)71 524 3091
Fax. +31 (0)71 526 6801
e-mail: P.J.H.de_Koning@lumc