Segmentation of Vessels in Peripheral CTA Datasets

Petr Felkel and Rainer Wegenkittl
VRVis Center, Austria, www.vrvis.at

Donau-City-Stra�e 1,  A-1220  Vienna
e-mail:

Armin Kanitsar
ICGA, Vienna University of Technology
Favoritenstra�e 9-11,  A-1040  Vienna
e-mail:


Contents


Abstract

The paper collects seven methods applicable in the task of peripheral vessel segmentation in 3D medical datasets, acquired by Computer Tomography Angiography (CTA) of the human leg. The fundamental aim of such a segmentation task is a robust method for the detection of main vessels in the leg that simultaneously preserves the vessel calcification (the sediment is called plaque) and allows localization of vessel narrowings (called stenoses). This segmentation has to be free from artefacts, i.e., without false detections of stenoses and without false omitting of any stenotic part. The methods described have been collected by a literature review focused on the vessel segmentation.

1  Introduction

The aim of the running project is to develop a fast and robust method for vessel and calcification segmentation, which requires a minimum of user interaction or concentrates the interaction into a minimal number of interventions. The input 3D datasets are acquired by Computer Tomography Angiography (CTA).

The aim of this paper is a literature research on the topic of vessel segmentation and information about the preliminary tests of interesting methods. It gives an overview of methods dealing with different kinds of vessel segmentation, i.e., not only with peripheral vessels and not only for CTA, as described later. The reasons are twofold: Firstly, only very few articles about CTA exist (or have been found by the authors), and secondly, the principles of 3D segmentation methods from different modalities can be adapted for the goal of the CTA segmentation. The studied methods include:

The presented text has the following structure: In Section 2, we characterize available acquisition techniques together with nine datasets available for testing. In Section 3, we describe the methods for vessel segmentation found up to now. In Section 4, we discuss experiences and potentials of each method for the task of peripheral vessels CTA segmentation. Conclusions follow in Section 5.

2  Data Description

The data acquisition methods applied and the datasets available for testing purposes are described in this section. To get a clearer picture of how the vessels are represented in the datasets more details about one of the datasets can be found in[7].

2.1  Why CTA when other exist?

Computer Tomography Angiography (CTA) data are used since the goal is: A non-invasive diagnostic method (in place of invasive DSA), detection and visualization of calcification with better spatial resolution (than MRA), shortening of current 4 hours manual vessel delineation (at first by an interactive method, later ideally by a fully automatic one), and measurements of vessel parameters. The principles, advantages and disadvantages of the modalities in more detail are as follows:

Digital Subtraction Angiography (DSA)
subtracts two X-ray images-one before and one after the local injection of a contrast agent into the patients vessels (using a catheter after a small surgery). It is an invasive and expensive method, with complication rates ranging from 0.17% to 7% [19]. Furthermore it is of 2D nature, which implies also limitations for vessels parallel to the viewing direction and limitations for detection and characterization of non-concentric stenoses. It does not show calcification.
Magnetic Resonance Angiography (MRA)
detects the electric signals from resonating protons in a strong magnetic field and reconstructs a full 3D volume. It scans the patient also after injection of a contrast agent, but at this time the contrast agent is applied non-invasively by a shot. MRA has a limited acquisition coverage of 40-48 cm (with a peripheral vascular coil up to 90 cm). It can not be used for patients with pacemaker, claustrophobia, metallic foreign bodies, and pregnancy. MRA gives in 98% the same results as DSA [19], but it does not show calcification.
Computer Tomography Angiography (CTA)
reconstructs a whole 3D volume from a large amount of X-ray projections. The main arguments for CTA are its (currently) better spatial resolution than CTA, better visualization of calcifications (which is the main goal), and the availability of CTA compared to MRA (and some contraindications of MR in general-see the previous item).

2.2  Problems of the CTA acquisition

In this section, we summarize the details of the data acquisition. The major problems of the data acquisition can be classified into three classes; homogeneity of the contrast agent distribution, settings of the CT scanner and partial volume artifact.

The so called plateau-like concentration of a contrast agent is necessary for getting a constant value response inside the studied vessels during the scanning. This means that the agent concentration should stay as constant as possible in the whole vessel of interest, which can be achieved by an appropriate contrast agent injection protocol and by a proper scan delay setting.

The plateau-like concentration of a contrast agent is typically studied and achieved in abdominal vessels. It is much more complicated to achieve the homogeneous concentration distribution in peripheral vessels, since the physical characteristics of small vessels is not known, and it becomes even more complicated if different amount of calcification, other types of stenoses and collateral flows exist. As an example of such a complicated concentration profile see Figure 5, where the profile in the abdominal part is relatively constant (right) but it changes rapidly in the lower parts of the body (left).

An important acquisition parameter, which can influence the quality of datasets and in this way also the resulting segmentation, is the setting of the CT scanner (collimation, gantry tilt, maximal dose, etc...). Modifications of these parameters can be tested after a properly-working segmentation method has been found to improve its reliability and accuracy.

Limited resolution of the scanner and mainly the scanning of slices thicker than the size of scanned tissue-structures result in a partial volume artifact (PVA) on tissue borders. If different tissues share a scanned voxel, the resulting scanned value is a weighted average of their values, which may be misclassified as a different tissue.

2.3  Example datasets

Table 1 contains the overview of nine datasets used for testing. The first five datasets are shown in a 3D view in Figure 1.


Figure 1: First five datasets in a 3D view

DS No  Slice   
No
of  sli- ces
Thick- ness
Spa- cing
Pixel sz x = y
Volume size
[mm]
[mm]
[mm]
[mm2x mm]
1
444
3    
2  
0.6484
3322 x 888
2
783
1.25
0.7
0.4629
2372 x 548
3a
553
3     
2  
0.6543
3352 x 1106
3b
169
3    
2  
0.25
1282 x 338
4
551
3     
2  
0.4688
2402 x 1102
5
1134
3     
1   
0.7422
3802 x 1134
6
1070
3     
1   
0.7266
3722 x 1070
7
988
2     
1  
0.5020
2572 x 1070
8
1166
3   
1  
0.7422
3802 x 1166

Table 1: Available datasets and their parameters

A note about the acquisition/reconstruction parameters: The columns thickness and spacing represent the parameters of the slices in each dataset (e.g., overlapping slices of 3 mm thickness with mutual spacing of 1mm). These parameters were given as input parameters to the reconstruction process at the CT scanner.

The acquisition parameters (4x2.5 or 4x1) describe the configuration of the scanner. This spiral-CT scanner acquires simultaneously four projection datasets and allows to configure the detectors to scan 2.5 or 1 mm thick projection datasets. From these raw projection-data the transversal slices are reconstructed with desired thickness and spacing. All the datasets in Table 1 have the acquisition parameters of 4x2.5 mm, except from No. 7 with 4x1 mm.

2.4  Stenotic vessel description

In this section the anatomical structure of a stenotic vessel and the appearance of the vessel and the surroundings tissue are described in an ideal cross-section above the knee.

A typical outer vessel shape is circular or slightly elliptical. If calcification is present, it is situated in the vessel and the inside diameter becomes considerably irregular (see Figure 2). Also non-calcified stenoses exist, formed by soft tissue deposits. Calcification is a second evolutional step of such a stenosis. The vessel dimension can be significantly enlarged if an aneurysm (a ball shaped enlargement of a vessel) is present.


Figure 2: Schematic cross-sectional anatomy of a diseased coronary vessel [24]

The stenotic vessel densities, measured by CT in Hounsfield Units (HU), should lie within these intervals [22]:

Adventitia
has the density equal to fat, i.e., -6510 HU.
Lamina and media
have the size comparable to the CT resolution (pixel size » 0.75mm) and due to partial volume artifact they are separately invisible. They modulate the voxel values of blood and calcification.
Vessel blood
without a contrast agent has 80±10 HU. With a contrast agent it should have 200 HU, but in our datasets the value varied from 150 to 260 HU.
Plaque
deposits show high densities (320 HU-1500 HU). There is typically more than one plaque piece in the vessel.

 


Figure 3: An example of voxel values in the cross section between knee and hip (artificially colored)

An example of the densities around the vessel above the knee is artificially colored in Figure 3. The vessel is surrounded by fat and muscles and can be separated from the bone very simply. But this is not the case for all vessels in the datasets, since the density values of blood in the vessels overlap with the densities of low-density bone and marrow (see the histogram in Figure 4).

 


Figure 4: Bone and blood vessel with overlapping densities (Courtesy M. Sramek, Austrian Academy of Sciences)

Some smaller vessels go directly into the bone and supply the inner bone structure with blood. This fact can complicate the vessel tracking approach.

Voxel densities along two vessels from dataset No. 8 were plotted in the following two graphs (see Figures 5 and 6). The vessel density distribution in a manually selected vessel in the left leg (see Figure 5) shows two large drop-downs in the density values, above the knee (slices 380-530) and in the upper part close to the hip (slices 685-741). These decreases of the contrast agent concentrations are caused by two stenoses, which drastically reduce the blood flow and therefore also the concentration of the contrast agent. The blood flow is spread into small collateral vessels. Below the large stenosis above the knee, blood returns to the vessel and the concentration grows back. In fact, it is even higher, since the contrast agent cumulates its concentration here. And also the veins become visible, as the contrast agent flow continues. The peaks in lower slices reflect an erroneously selected voxels with calcification.

 


Figure  5: Topogram image of dataset No. 8 with a dark line indicating the manually segmented vessel of the left leg and the voxel density values along the vessel together with the average values of density from a 3x3 surrounding of the center-path

 


Figure 6: CMPR of another vessel of the dataset No. 8 and the graph of voxel density values along this vessel
(segmentation by the tracking algorithm from Section 4.6)

3  Vessel Segmentation Methods

In this section we summarize the knowledge obtained from the literature review on the topic of vessel segmentation. The goal is a robust method for the vessel segmentation from CTA datasets. Since only one technical article using this modality for a vessel display was found, this section collects everything about segmentation of vessel objects obtained until now, describes the principles and tries to depict problematic steps of each method. Most of the methods described in studied articles deal with DSA (digital subtraction angiography) and MRA (magnetic resonance angiography).

For details about the methods read this section, for discussion about their applicability in the task of 3D vessel segmentation see Section 4.

3.1  Threshold-morphological method

The outline of this section results from personal communications with Milos Sramek from the Austrian Academy of Sciences. As already mentioned in Section 2.4, both tissues of interest, namely vessels and calcifications, overlap in their CT density range with bone tissue. Therefore, a technique capable of independent labeling of all three tissue types is necessary. In general, two complementary strategies can be followed: An indirect one, in which the bone tissue is identified and removed from the CT data by means of masking. In this case vessels and calcifications remain the tissues with the highest voxel density, which enables a straightforward application of subsequent processing and visualization techniques. And a direct one, in which the vessels and calcifications are segmented and labeled from the original data.

Sramek prefers the indirect approach since bones are much thicker objects than vessels and, therefore, their segmentation is less sensitive to imaging artifacts (partial volume effect), and bone segmentation errors do not have so dramatic consequences as the segmentation of vessels can have (e.g., missing vessel segments).

According to the nature of the CTA data, Sramek recommends a thresholding, supplemented by morphologic operations and a connected component labeling as the most promising segmentation technique. The tools must be developed to deal with situations (rare but possible), when the simple operations are not powerful enough to accomplish the bone segmentation correctly.

The data with masked bone tissue can be further used for the segmentation of vessels and/or calcifications (thresholding, statistical techniques, region growing, vessel tracking) and for visualization (MIP-maximal intensity projection, re-projection, direct volume rendering). Since now, after masking of the bone tissue, the vessels and calcifications have the highest density within the volume, these tasks are significantly simplified.

3.2  Wave vessel tracking

The application area of Wave algorithm is vessel tracking in CTA, MRA

Cornelia Zahlten described the Wave vessel tracking algorithm in her PhD thesis [30]. She mentioned two approaches and used the second one:

I. Hysteresis thresholding with skeletonization and symbolic graph [11,25]. Object pixels are selected by hysteresis thresholding: The histogram is divided into 3 parts-object, background, and overlapping part. Pixels from the overlapping part are added to the object only if they are connected to object pixels. The object is then skeletonized by erosions and the distance transformation is then computed to determine the diameter of the vessels. Finally, a symbolic description via a graph is constructed, where the vessel bifurcations and vessel ends are represented as nodes and the vessel segments as links of the graph.

II. Region growing with simultaneous graph generations [30]. This is a region growing in waves approach enriched by bifurcation detection and vessel graph generation. At first, a seed point is put interactively near the root of the vessel tree. Then, the wave goes through the object. The wave is a connected list of voxels belonging to the vessel that have been added in the current step. Until now it is a normal flood-fill algorithm with 26-connectivity voxels, each voxel remembers where the wave has come from (26 binary labels) and the wave direction correction can be applied for distorted vessels.

The major difference to the classical region growing is in the bifurcation detection and the simultaneous graph construction. If the voxels in the wave (newly added voxels in current step) are not mutually connected, the bifurcation (n-furcation) is detected and an appropriate number of nodes (one for each branch) is created and inserted to the graph. The new wave parts are sorted according to the number of voxels and processed from the largest to the smallest one. This results in the construction of the vessel tree in a top-to-bottom order. Each new branch gets a new label, except for the largest one (75% of the current wave), which holds the current label. During the visualization step each label is assigned a different color to show the hierarchy in the vessel tree.

For a correct branch detection (to avoid over-branching in places with a high vessel curvature), the wave correction is necessary to assure the next wave step would be perpendicular to the actual vessel direction. Details in [29, 30].

Zahlten discusses also a data preprocessing step-a median filtering, which eliminates thin vessels, or a possibility to use an adaptive filtering by local region growing, which preserves thin connections. The adaptive filtering is done iteratively. The possible best filtering method, which eliminates differences in contrast agent distribution, is the (time consuming) combination of both filters, i.e., an application of the adaptive regional growing after median filtering.

3.3  Vessel enhancement and tracking

The application area is vessel segmentation of abdominal and carotid arteries in MRA.

The principle of this three stage algorithm ([8, 9, 10], similar approaches [14, 20, 21, 26]) is at first to do a preprocessing of the dataset by a 3D filter (vesselness operator V), which enhances tubular structures, then to track the tubular structures in 3D space by finding the path with a minimal cost, which goes through the centerline of the vessels, and finally to detect the vessel borders. The vesselness operator uses secondary derivatives for the detection of the principal object shape and works in a scale space [15], i.e., at each voxel, a set of filterings is done with different sizes of the vesselness operator and the maximal response (which appears when the size of the operator matches the size of the vessel) depicts the vessel size.

This preprocessing step is applied to enhance 3D tubular structures in the dataset. The vesselness enhancement filter V of Frangi et al. [9] is based on the eigenvalues of a Hessian matrix (a matrix of partial secondary derivatives) evaluated in each pixel of the image in different scales s.

For each voxel, the filtering is done in different scales s �smin, smax as follows: After computation of the eigenvalues, they are ordered so that |l1| < |l2| < |l3|. The respecting eigenvectors  u1, u2 and  u3 then point out the singular directions:  u1 along the vessel, and  u2u3 the directions in an orthogonal plane. The value of the vesselness operator V(x,s) is than computed and the highest response over the scales is reported as the response of the vesselness filter V(x). The scale s where V(x,s) reaches its maximum determines the size of the vessel with center in voxel x.

The value of vesselness operator V(x,s) for a given scale s in voxel x is for brighter vessels in the darker environment defined as:

The operator consists of three parts normalized to have the response in interval 0,1:

Parameters a,b,c tune the sensitivity of the filter to deviations in RA, RB and S. Typical values used in [9] are a = b = 0.5 and c = one half of the maximal Frobenius norm of the Hessian matrices, i.e., one half of the length of the vector of lambdas (l1, l2, l3). Another possibility is to use a normalization by the maximal intensity 0.25s2Imax.

The operator enhances the tubular structures, filters out blobs and plateau-like structures, but also reduces the vessel diameter, has dropouts at bifurcations (detects them as to be plate-like) and is highly sensitive to bones and calcifications in CTA datasets.

After filtering by a vesselness operator, the vessel centerline is detected by 3D snakes and modeled by B-spline curves, then the vessel wall is segmented and modeled by a tensor product B-spline surface.

3.4  Direct vessel tracking in 3D

The application area is CTA of abdominal aorta and MRA.

Wink et al. [28] described an interactive segmentation method, which works locally, without preprocessing of the whole dataset and needs about 10s per vessel. The user selects two start-points A and B in the thick part of a vessel and runs the tracking algorithm. The points give a possible direction of the vessel centerline a = B-A. The method estimates the position of the next candidate point C in this direction of vector a, in the distance b based on the current minimal vessel diameter dmin. Precisely b = admin, where a is a given constant.

Then the precise position of a new point Cnew is computed. In the square surroundings of the candidate point C, which is perpendicular to the vessel axis direction a, they compute for each point the likelihood that it is a center point of the vessel. This is done by casting of a fan of rays from each point in the square. Each ray ends at voxel that locally looks as the vessel border. Then, they detect the most possible vessel center point Cnew in this square and store this new position as a new centerline end. It becomes the part of up-to-now detected vessel and a new candidate point is generated in the computed axis direction. The estimated point C need not to lie inside of the vessel-it is sufficient when the vessel center lies in the square around it [28, Fig. 8]. The size s of this square is computed as s = bb, where b is another given constant.

The center likelihood is computed as follows: For each pair of rays in opposite directions the likelihood of being-a-center is computed as a ratio of the shorter to the longer of the distances to the vessel border. The border is detected as a falling gradient in the same direction as the direction of the ray. This has an advantage since using this approach the border of bright vessel or the end of calcification can be found.

The authors also proposed a possibility to force a search direction to the central vessel direction, to limit the curvature of a vessel by a coefficient and to search a whole tree of possible vessels and continue in the direction of the highest sum of center likelihoods. They also proposed an approach, which is based on the same principle as our modification to the live wire method (see Section 4.6)-namely to define start- and end-points and to find the vessel centerline by the use of the dynamic programming or tree search. The more points are defined, the more the searched data space is reduced. A similar tracking approach called imaginary catheter was published as a work in progress by Verdonck et al. [27].

3.5  Real time vessel enhancement

The application area is real-time segmentation of DSA.

The principle of this method published by Poli and Valli [17, 18] is similar to the approach of Frangi et al. (see Section 3.3). Both use Gaussian filters in different scales for vessel enhancement. Cagnoni et al. [1, 2] applied then a genetic algorithms optimization. The main points of Poli and Valli's algorithm are:

To reduce the filter artifacts, the authors decompose the directional derivatives not only in two directions but in four vector n = [1,0], [0,1], [(2)/2], [(2)/2], [(2)/2],-[(2)/2] and as a result, they take a maximum response over all n,w,l. According to their test image, where the best parameter setting was s = 1, w = 2, l = 2, it seems that no integration over set of scales is necessary!

3.6  Live wire and Intelligent scissors

The application area is a boundary segmentation of any object.

The live wire (LW) method [5, 6] was originally designed for a 2D interactive segmentation of boundaries in difficult images with faint signal of boundaries or even gaps in boundaries, for objects with similar boundary properties around them, and for regions distorted by noise. In general, it speeds-up the user interaction by interactive offering the boundary segments between an user defined start-point and the current position of the cursor. After selection of the end-point of the segment, it becomes a new start-point and the whole process of boundary segment selection starts from this new point.

The fundamental ideas of this method and the name live wire were developed in cooperation of groups around Udupa and Barrett. But then (1995/96), the development continued independently and Mortensen and Barrett [16] introduced the name intelligent scissors. The important principles used by Falcao et al. [5] are:

The boundary is an oriented, closed and connected planar contour. It is formed from a subset of oriented edges such that the sum of the costs of these edges is minimal. The oriented pixel edge is defined between each pair of four-adjacent pixels. The oriented pixel edge element, which is a part of the boundary, is called bel, a boundary element. Each bel b = (q,r) has its location (between pixels q and r) and orientation (such that q is always inside the boundary). Each bel b is assigned a cost(b), computed from the selected set of bel features. Details about the computation of the costs, about the possible training phase and also the explanation of the relation to intelligent scissors [16] are given in [6]. Pixels and bels form nodes and arcs of a weighted directed graph. A modified Dijkstra's minimal cost path algorithm is used.

For speeding up of the interaction times, Falcao et al. [5] proposed a method called live wire on the fly (LWOF), which computes the paths ``on demand'' according to the cursor position.

3.7  Knowledge-based 2D approach

The application area is segmentation of vessels in DSA.

The development of a hierarchical heuristic method for a knowledge-based vessel segmentation from projections in subtraction angiography (SA) was started by Cleynenbreugel et al. [3], and continued in [4, 23]. The method applies a rule-based system and segments the blood vessels via a multi-stage approach.

The proposed multi-layered image representation [23] has the following layers:

Pixels
with their position and gray value. They form the bottom level of the hierarchy.
Ridge-point segments
are the high intensity line-structures after thinning and linking.
Blood vessel segments
are parts of vessels detected as ribbon-like structures around extracted centerlines. A new vessel segment is created at each vessel crossing.
Blood vessels
are created by merging the blood vessel segments. Assumption of a tree-like structure is utilized, models of crossings, connections and T-branching are applied. Intersections are labeled according to geometrical and topological knowledge.
List of anatomically labeled blood vessel segments
is created by interpretation of vessel segments according to the anatomical model of a coronary artery. The anatomical labeling may overrule the available geometrical knowledge in some cases.
3D artery structure
is finally reconstructed from the available multiple DSA projections.

An important aspect is the splitting of large amount of available heuristic knowledge into separate blocks, which simplifies integration of new heuristic models. Also the feedback strategy, which allows re-computation of missing attributes and the updating of interpretation labels makes the method more robust.

4  Applicability of the Methods

In this section, we describe our experience with the algorithms described in Section 3 and collect our opinions about the applicability of described methods to the task of vessel segmentation in CT angiography datasets. Up to now, four algorithms were tested: Threshold-morphological method (Section 4.1), wave vessel tracking (Section 4.2), 2D version of vessel enhancement (Section 4.3), and a live wire (Section 4.6). Therefore, about the other methods only opinions are mentioned, which are not supported by precise algorithm evaluation.

4.1  Tests of threshold-morphological method for bone removal

The bone removal is very useful mainly for the visualization using a maximum intensity projection. If used as a preprocessing step for vessel segmentation, incorrectly removed vessel parts may appear, caused by the applied thresholding. The following vessel segmentation algorithm may then suffer from false detected stenoses and from a problem of jumping over gaps in the data. The distort vessel shape would also affect the following measurements.

A combined bone segmentation and removal method, which uses a two-phase thresholding and region growing followed by average-value and region-size based labeling, was tested with very promising results [12]. The method works in slabs to achieve interactivity. The method cannot handle variations of intensities in the bone automatically and has to allow a user interaction, as careful settings of the segmentation parameters is still often necessary.

4.2  Tests of vessel tracking by Wave

A simplified version of this approach with an adaptive thresholding by means of bimodal histogram without a curvature correction was implemented. Before application of this algorithm, it is necessary to enhance the contrast between the vessel and the background by setting of a correct windowing (to select only an interval of data values).

The algorithm was tested at the beginning of the bibliography search. A two-step segmentation was used: At first, the vessel surroundings was masked out by the interval selection and then the vessel tracking itself was performed. The masking of the volume had the comparable effect as the process of windowing.

The algorithm was successful for the parts of the leg starting below the knee and ending at the lower part of the body if the windowing/masking isolated all the bridges to the surrounding tissues (The simplest dataset was No. 3a which has higher resolution-see two right panels in Figure 7). The algorithm stopped very often in small vessels in the lower part of the foot as the vessel was no more homogeneous enough due to the noise and may be the partial volume artifact (see Figure 7 left). Also if the vessel touched the bone in the abdomen or if a star-like reconstruction artifact was present, it spread into the surrounding tissue or even filled up the whole 3D dataset. Another important disadvantage was a non-predictable erosion of the detected vessel caused by ignoring the voxels out of the windowing interval.


Figure 7: Datasets No. 1 (left) and 3a (right) and segmented vessels

We suspect that the windowing, which is necessary to achieve any meaningful result, can suppress the adaptive thresholding to minimal influence and can cause an unpredictable erosion of the detected vessel. The detection of small vessels is then very hard (see limited length of vessels in the left panel in Figure 7) and the algorithm without branching corrections reduces itself back to the region growing. Also due to a very low interactivity and many fails it was realized not to continue in testing and not to use this algorithm for the project.

4.3  Test of vessel enhancement

The algorithm has been developed for MRA vessel segmentation where no signal from bone is present, and especially for abdomen, where the vessels have a large diameter. As bone is present in the CTA datasets, the approach cannot be applied directly.

But the 3D filtering step, which enhances tubular structures of diameter from a given interval and even detects the vessel diameter, can be used in two places: In the modification of the live wire vessel tracking (see Section 4.6) as a supporting feature helping in the detection of "being in the vessel", or as a post-processing step for a precise positioning of the detected path (e.g., the minimal path) to obtain a true centerline of the vessel.

For large diameters the vesselness method is precise enough (carotid arteries). For small vessels of diameter of 2-3 voxels (like in brain or foot), the Hessian is not computed in the correct centers of the vessels, but in the centers of sampled voxels, and the variation of anisotropy term RA = |l2| / |l3| is therefore too high [14].

2D tests showed that the operator enhances the vessels of the selected diameter. But it also turned out that the image has to be extended to omit false edge responses on the borders, which slows down the filtering.

Frangi et al. used snakes for the vessel extraction from the vesselness data. It should prefer detection of smooth vessel centerlines, i.e., should work globally in a local neighborhood. This may be useful for jumps over bifurcations or calcifications.

4.4  Direct vessel tracking

This method is the only one in this paper that was directly applied to CTA data. It seems to be very interesting, as it does not search the whole data volume but only the promising part, where the vessel should be located. It also handles the calcifications in CTA with a correct estimate of the outer border of calcifications, ignores perpendicularly outgoing vessels as they have a low contribution to the likelihood, selects one vessel in bifurcations (no wrong turnings or leaving of the vessel), and estimates the vessel diameter. It is invariant to intensity, scale and rotation.

On the other hand, the method was used for vessels with a large diameter in abdomen, which is not our case. Also the necessity of permanent user interaction and the start points selection may be a drawback in the large datasets. It needs a simultaneous visualization and a specialized user interface.

4.5  Real-time vessel enhancement

This algorithm differs from the Frangi's vessel enhancement by application of different derivative kernels, with different width w and length l. The authors also use integer arithmetic and a sophisticated decomposition of the kernels into addition of a large number of responses of only 2-pixel kernels, which results in a real-time vessel segmentation.

Surprisingly, Poli et al. do not use the kernels over the range of scales s but only for s = 1 !!! Also the best compromise values of s = 1, w = 2, l = 2, seems to be relatively strange, as they negate the scale-space approach.

The ideas of optimal computation of the convolutions are inspirative and should be in mind if any filtering approach will be applied.

4.6  Tests of modified Live wire

We have proposed a modification [12] of the original algorithm in order not to search the boundary but the path through a homogeneous region in a given window of values. That means the path through the vessel (see Figure 6 for an example of such a path projection).

The proposed implementation needs an interactive selection of a start-point and a set of end-points: the start-point in the abdominal aorta and the end-points at the ends of the tracked vessels. Then, the shortest path search between the start-point and all the points in the image (graph) is performed. The algorithm prefers homogeneity (smallest difference of values), allows jumps to higher densities (calk), but jumps out of the given window are highly penalized (to prevent tracking in the false, but homogeneous regions of different tissues). The algorithm works very intuitively, the user interaction is concentrated to the beginning of the session, which saves time of the operator. After the processing of the dataset, the paths between start and all endpoints are drawn. At this point, any shortest path from the start-point to any other end-point can be found interactively.

The proposed algorithm modification was tested on 2D and 3D datasets with promising results. The algorithm detects vessels with a higher probability of success. On the contrary, a non-optimized version needs a large amount of memory, which is now solved by swapping to disk. Also a modification of any point inside the vessel means the setting of this point as a starting one and the complete re-computation of the shortest path (sub-)tree. The graph parameters are fixed, i.e., no tuning according to the dataset via a teaching phase is supported.

4.7  Knowledge-based approach

This approach was not tested, i.e., only some notes follow.

Cleynenbreugel/Smets processed the 2-dimensional DSA datasets and used a thin-vessel detection by the maximal image-intensity detector. This detector is not applicable to CTA, as the whole 3D dataset is processed and no 2D MIP of vessels exists. The approach seems to be robust, as it combines a bottom-up approach with a feedback based recalculation and re-labeling and applies not only the geometrical but also the anatomical knowledge. On the other hand, segmentation of the whole 3D CTA datasets is more simple as it does not need to deal with overlapping and crossing of vessels caused by a 2D projection of a 3D object (as in DSA).

5  Conclusions

The paper summarized the information about angiography datasets acquired by a spiral CT scanner and about the appearance of vessels of interest in these datasets. Then it described seven methods for segmentation of vessels in medical datasets found in literature and discussed their applicability to the task of CTA vessel segmentation.

The thresholding-morphological method was not usable for vessel segmentation, but was advantageously applied for bone removal for MIP visualization. Wave vessel-tracking was too slow for large datasets with lower reliability in comparison to the shortest-path search between given pairs of points. The vessel enhancement method enhanced bone more than vessels in CTA datasets. But it can be used as a supporting feature for vessel centerline detection. The principles used in direct vessel tracking were used for vessel centerline detection and not for vessel tracking itself, mainly as the lower extremity arteries have a substantially smaller diameter and also the user interaction cannot be concentrated to the beginning of the session. We assume that the real-time vessel enhancement would work similarly as the vessel enhancement method. The knowledge based method was too specialized for 2D case, which is not necessary in case of CTA. The most promising approach is the modification of the live wire method for vessel detection.

After the tests, we have decided to use two methods; a bone removal for MIP projection generation and the combination of approaches of live wire and direct vessel tracking for vessel and its centerline detection. Preliminary results have already been published in [12] and [13]. In the future we will concentrate on tuning of the implemented methods and tests in the clinical environment.

Acknowledgments

This work was done in the VRVis Research Center, Vienna (http://www.vrvis.at) and in TIANI MedGraph AG, (http://www.tiani.com). Datasets courtesy D. Fleischmann, AKH Vienna. Special thanks to E. Groeller, D. Fleischmann, D. Sandner, M. Sramek and M. Capek for valuable discussions, and to the referees, whose comments considerably improved the presented text.

References

[1]
S. Cagnoni, A. Dobrzeniecki, R. Poli, and J. Yanch. Genetic algorithm-based interactive segmentation of 3D medical images. Image and Vision Computing, 17(12):881-896, 1999.
[2]
S. Cagnoni, A. B. Dobrzeniecki, R. Poli, and J. C. Yanch. Segmentation of 3D medical images through genetically-optimized contour-tracking algorithms. Technical Report CSRP-97-28, University of Birmingham, School of Computer Science, Dec. 1997.
[3]
J. V. Cleynenbreugel, F. Fierens, C. Smets, P. Suetens, and A. Oosterlinck. Knowledge-based segmentation of subtraction angiograms. In IPMI'87, pages 307-314, Utrecht, The Netherlands.
[4]
D. Delaere, C. Smets, P. Suetens, G. Marchal, and F. V. de Werf. A knowledge-based system for the 3D reconstruction of blood vessels from two angiographic projections. In J. Cornelis and S. Peeters, eds., Proceedings of North Sea Conference on Biomedical Engineering, Belgium, 1990.
[5]
A. X. Falcao, J. K. Udupa, and F. K. Miyazawa. An ultra-fast user-steered image segmentation paradigm: Live wire on the fly. IEEE Transaction on Medical Imaging, 19(1):55-62, Jan. 2000.
[6]
A. X. Falcao, J. K. Udupa, S. Samarasekera, S. Sharma, B. E. Hirsch, and R. A. Lotufo. User-steered image segm. GMIP, 60:233-260, 1998.
[7]
P. Felkel. Segmentation of vessels in peripheral CTA datasets. VRVis Research Center Technical Report TR-VRVis-2000-008, Vienna, Austria, Dec. 2000.
[8]
A. Frangi, W. Niessen, R. Hoogeveen, T. van Walsum, and M. A. Viergever. Model-based quantitation of 3D MR angiographic images. IEEE Transaction on Medical Imaging, 18(10):946-956, 1999.
[9]
A. Frangi, W. Niessen, K. Vincken, and M. Viergever. Multiscale vessel enhancement filtering. In W. Wells, A. Colchester, and S. Delp, editors, MICCAI'98, volume 1496 of LNCS, pages 130-137. Springer-Verlag, Germany, 1998.
[10]
A. F. Frangi, W. J. Niessen, P. J. Nederkoorn, O. E. H. Elgersma, and M. Viergever. 3D model-based stenosis quantification of the carotid arteries from contrast-enhance MRA. In Mathematical Methods in Biomedical Image Analysis, 2000. In press.
[11]
G. Gerig, T. Koller, G. Szekely, C. Brechbuhler, and O. Kubler. Symbolic description of 3-D structures applied to cerebral vessel tree obtained from MR angiography volume data. Lecture Notes in Computer Science, 687:94-111, 1993. Cited according to [30].
[12]
A. Kanitsar, R. Wegenkittl, P. Felkel, D. Sandner, E. Groller, and D. Fleischmann. Automated vessel detection at lower extremity multislice CTA. In European Radiology, 2001, volume 11(S1), page S236. Springer Verlag, Heidelberg, Germany, 2001. Presented at ECR 2001, Vienna, Austria.
[13]
A. Kanitsar: Advanced Visualization Techniques for Vessel Investigation, Masters Th., TU Vienna, 2001.
[14]
K. Krissian, G. Malandain, N. Ayache, R. Vaillant, and Y. Trousset. Model based detection of tubular structures in 3D images. Research Report RR-3736, INRIA, Sophia Antipolis, July 1999.
[15]
T. Lindeberg. Scale-space theory: A basic tool for analyzing structures at different scales. J. of Applied Statistics (Supplement on Advances in Applied Stat.: Statistics and Images: 2), 21(2):224-270, 1994.
[16]
E. N. Mortensen and W. A. Barrett. Intelligent scissors for image composition. In SIGGRAPH'95., pages 191-198, Los Angeles, USA, Aug., 1995.
[17]
R. Poli and G. Valli. Efficient enhancement and segmentation of blood vessels by integration of multiple directional filters. Technical Report 940701, Dept of Electronic Eng., University of Florence, Italy, 1994.
[18]
R. Poli and G. Valli. An algorithm for real-time vessel enhancement and detection. Computer Methods and Programs in Biomedicine, 52:1-22, Nov. 1996.
[19]
S. G. Ruehm, T. F. Hany, T. Pfammatter, E. Schneider, M. Ladd, and J. F. Debatin. Pelvic and lower extremity arterial imaging. AJR, 174:1127-1135, 2000.
[20]
Y. Sato, S. Nakajima, H. Atsumi, T. Koller, G. Gerig, S. Yoshida, and R. Kikins. 3D multi-scale line filter for segmentation and visualization of curvilinear structures in medical images. In Proc. First Joint Conference of CVRMed II and MRCAS III, pages 213-222, Grenoble, France, 1997.
[21]
Y. Sato, C.-F. Westin, A. Bhalerao, S. Nakajima, N. Shiraga, S. Tamura, and R. Kikins. Tissue classification based on 3D local intensity structures for volume rendering. IEEE Transaction on Visualization and Computer Graphics, 6(2):160-180, 2000.
[22]
E. Schuster. Bimedizinische Technik. Uni-Wien, 1996.
[23]
C. Smets. A knowledge-based System for the Automatic Interpretation of Blood vessels on Angiograms. PhD thesis, Katholieke Univesiteit Leuven, 1990.
[24]
M. Sonka, V. Hlavac, and R. Boyle. Image Processing, Analysis and Machine Vision. Published by PWS Publishing, Brooks/Cole, CA, 2 edition, 1999.
[25]
G. Szekely, T. M. Koller, R. Kikinis, and G. Gerig. Structural description and combined 3-D display for superior analysis of cerebral vascularity from MRA. In Visualization in Biomedical Computing, volume 2359, pages 272-281, 1994. Cited according to [30].
[26]
D. Vandermeulen, C. Smets, P. Suetens, G. Marchal, J. Gybels, and A. J. Oosterlinck. Knowledge-based 3-D segmentation of blood vessels on a spatial sequence of MRI and ultrasound images. In S. J. Dwyer, R. G. Jost, and R. H. Schneider, editors, Medical Imaging III: Image Processing, volume 1092 of Proc. SPIE, pages 142-148, 1989.
[27]
B. Verdonck, I. Bloch, H. Maitre, D. Vandermeulen, P. Suetens, and G. Marchal. Blood vessel segmentation and visualization in 3D MR and spiral CT angiography. Computer Assisted Radiology, pages 177-182, 1995.
[28]
O. Wink, W. Niessen, and M. Viergever. Fast delineation and visualization of vessels in 3-D angiographic images. IEEE Transaction on Medical Imaging, 19(4):337-346, Apr. 2000.
[29]
C. Zahlten. Reconstruction of branching blood vessels from CT-data. In M. Gobel, H. Muller, and B. Urban, editors, Visualization in Scientific Computing, pages 41-52. Springer-Verlag, Wien, 1995.
[30]
C. Zahlten. Beitrage zur mathematischen Analyse medizinischer Bild- und Volumendatensatze. PhD thesis, Universitat Bremen, 1995.