| Author(s) | Graichen, U. |
| Title | Rekonstruktion der Koronar-Anatomie mittels digitaler Bildverarbeitung von Echokardiogrammen |
| Abstract | One of the main causes of death in the western countries are coronary heart
diseases. Presently, the X-ray angiography is the gold standard for the
examination of coronary arteries. This invasive, X-ray-based procedure
requires the infrastructure of a cardiac catheterization lab. Ultrasonic
examinations are getting more and more important as an imaging technique in
cardiography. In contrast to the X-ray angiography, patient and physician are
not exposed to degrading radiation. Further, ultrasound devices are commonly
used already and inexpensive in comparison to other imaging systems. The goal
of this project was to develop an image processing system, which can detect
and quantify coronary arteries in 3D-ultrasound data and thus can supplement
or replace the X-ray angiography examination. The procedure consists of
several steps: 1st the interpolation of volumedata without artefacts, 2nd the
analysis of the local structure of the datasets, 3rd tracking the run of the
vessels and 4th the segmentation of the vessel lumen.
3D ultrasound datasets are interpolated out of a series of 2D ultrasound images, which are recorded using a transesophageal transducer. The recording is triggered by the ECG and needs a few minutes. Because of the long recording time, the motion of the heart and of the patient during the recording the image series contain movement artefacts. This artefacts can be reduced by registration adjacent images in the series before interpolation volume data. The registration is done using a rigid, correlation based method. Rotation and scaling parameter are determined from the Fourier-Mellin invariant descriptors of the images. The translation parameters are calculated out of the by rotation and scaling corrected images. Due the special structure of the ultrasound images a windowing function adapted to the ultrasound cone is used. With this modified volume interpolation method movement artefacts can be significant reduced. The analysis of the local structure is used to detect line-like structures in 3D ultrasound datasets. The structural analysis is done using the local differential structure of the datasets. Particularly suitable for this purpose are Hessians which are calculated for every voxel of the Gaussian-smoothed 3D dataset. The Eigenvectors of the Hessian establish an orthogonal system. Two Eigenvectors are pointing in the directions with the minimal or maximal absolut value of the second derivation respectively. The third is orthogonal to both. The Eigenvalues connected to the Eigenvectors spezify the value of the second derivation. From the Eigenvalue a similarity measure with Gaussian lines is calculated. By chosing the the standard deviation and by examination the sign of the Eigenvalues it is possible to search selective of a line-like with a specific diameter and progression of contrast. The run of the vessels is described by centerlines. The result of the structural analysis step is 3D dataset with similarity measures with the select line-like structure. The centerlines are calculated in such a way that the run of the centerline is close to voxels with large similarity measures. This done by a modified thinning method. First a threshold is applied. Voxels with a similarity measure smaller than this threshold are set to zero. From the remaining voxels a erasure list is generated. The erasure list contains coordinates of the voxels and is sorted by the similarity measures in ascending order. During the thinning the erasure list is traversed and it is checked to delete the voxel without changing the topology and without removing end points of the skeleton. The resulting resulting skeleton runs close to voxels with large similarity measures. Along the centerlines of the vessels the lumen of the vessels is segmented. For every voxel on the centerline a cross section is calculated. The normal of the cross section is the Eigenvector of the Hessian which is connected to the Eigenvalue with the smallest absolute value. On the cross sections the lumen of the vessels is segmented using a modified Snake-method. The potential function, which describes the energy of the image, is calculated from the magnitude of the gradients and additionally from the similarity measure with line-like structures. The potential forces which affect to the Snake are calculated from this potential function using the gradient-vector-flow method. By this modified Snake method is robust approach for segmenting the lumen of vessels. |
| Download | Graichen05.pdf |