DOI: 10.1148/rg.251045051
RadioGraphics 2005;25:263-271
© RSNA, 2005
Informatics in Radiology (infoRAD)
Three-dimensional Atlas of the Brain Anatomy and Vasculature1
Wieslaw L. Nowinski, DSc, PhD,
A. Thirunavuukarasuu, MSc,
Ihar Volkau, PhD,
Rafail Baimuratov, MSc,
Qingmao Hu, PhD,
Aamer Aziz, MD, PhD and
Su Huang, MSc
1 From the Biomedical Imaging Lab, Bioinformatics Institute, 30 Biopolis St no. 07–01, 138671 Singapore. Recipient of a Certificate of Merit award for an infoRAD exhibit at the 2003 RSNA Scientific Assembly. Received March 24, 2004; revision requested June 17 and received July 6; accepted July 21. Supported by the Biomedical Research Council of the Agency for Science, Technology and Research, Singapore. All authors have no financial relationships to disclose. Address correspondence to W.L.N. (e-mail: wieslaw@bii.a-star.edu.sg).
 |
Abstract
|
|---|
Of the existing atlases of the brain anatomy and cerebrovasculature, none integrates the anatomy and vasculature by providing for direct manipulation of three-dimensional (3D) cerebral models. An atlas-based application was developed in four steps: (a) construction of 3D anatomic models, (b) construction of 3D vascular models, (c) interactive spatial coregistration of the anatomic and vascular models, and (d) development of functionality and a user interface for the application. Three-dimensional anatomic models were imported from an electronic brain atlas database derived from classic print atlases. A novel vascular modeling technique was developed and applied to create a vascular atlas from magnetic resonance angiographic data. The use of 3D polygonal models allows smooth navigation (rotation, zooming, panning) and interactive labeling of anatomic structures and vascular segments. This application enables the user to examine 3D anatomic structures and 3D cerebral vasculature and to gain a better understanding of the relationships between the two. The combined anatomic-vascular atlas is a user-friendly neuroeducational tool that is useful for medical students and neuroscience researchers as well as for educators in preparing teaching materials.
© RSNA, 2005
|
Introduction
|
|---|
Numerous electronic atlases of the brain have been developed as research prototypes, public domain tools, and commercial products, such as The Digital Anatomist (Structural Informatics Group, Department of Biological Studies, University of Washington, Seattle) (1), Voxel-Man (Institute of Medical Informatics, University Hospital Hamburg-Eppendorf, Hamburg, Germany) (2), and Cerefy (a joint effort of Thieme [New York] and Biomedical Imaging Lab [Singapore]) (3–6). These atlases have been reviewed in previous articles (7–9). A substantial effort in building digital atlases of the brain has been undertaken by the International Consortium for Brain Mapping (10–13).
Unlike existing brain atlases, the majority of which were developed mainly for educational and research purposes, the Cerefy family of brain atlases is targeted for clinical, research, and educational purposes (14). Ten Cerefy applications developed in our research laboratory are being used worldwide in stereotactic and functional neurosurgery (5,8,15,16), neuroradiology (17,18), human brain mapping (3,19,20), neuroeducation (4), and research (14). Two Cerefy libraries, the Cerefy Electronic Brain Atlas Library and the Cerefy Brain Atlas Geometrical Models (6,8,21), are integrated with the major imaging-guided surgical workstations, including those of Medtronic (Minneapolis, Minn), BrainLAB (Heimstetten, Germany), and Elekta (Stockholm, Sweden). The Cerefy Clinical Brain Atlas: Extended Edition with Preoperative Planning and Intraoperative Support (5) is a stand-alone tool that provides anatomic and functional atlases for neurosurgical planning. An algorithm for fast calculation of the probabilistic functional atlas (9) is the core component of a public domain portal for stereotactic and functional neurosurgery (available at www.cerefy.com) (22). The Brain Atlas for Functional Imaging (3) is a tool for atlas-based localization analysis of functional images. The Cerefy Neuroradiology Atlas (also available at www.cerefy.com) (17) facilitates atlas-based analysis of multiple data sets. The Cerefy Atlas of Brain Anatomy (4) is a user-friendly tool for exploring and learning about the anatomy and for creating teaching materials. This atlas is also available for medical image analysis research in ANALYZE (Analyze Direct, Lenexa, Kan).
Several electronic atlases available on CD-ROM demonstrate the cerebral vasculature (2,23–25). These atlases typically contain superb content but allow only limited navigation. Because precalculated three-dimensional (3D) images are used, smooth rotation and zooming are not feasible. For instance, Voxel-Man (2) demonstrates 3D blood vessels derived from magnetic resonance (MR) angiographic data. Because Voxel-Man makes use of precalculated 3D and maximum-intensity-projection images, interaction with the 3D images is achieved with 3D Quick Time (Apple Computer, Cupertino, Calif) movies. Similarly, the atlas developed by Kretschmann and Weinrich (24), with high-quality content derived from cryosections, contains precalculated 3D images of the brain and cerebral vasculature.
Our goal is to develop a combined anatomic-vascular interactive brain atlas whose initial application is in neuroeducation. A 3D version of the Cerefy brain atlas was used to obtain the subcortical anatomy, whereas the vascular part of the atlas was constructed from angiographic data. The combined atlas does not have the limitations of existing vascular atlases, since the user can freely manipulate 3D models by rotating, moving, and panning them in real time. In addition, all anatomic structures are labeled with their names. Any structure can be selected from the index for display or can be labeled by pointing to it with the cursor. Our atlas enables the user to study and learn about 3D anatomic structures and 3D cerebral vasculature and to gain an understanding of the relationships between the two.
In this article, we review the materials and methods used in the construction of the anatomic, vascular, and combined atlases. We also discuss and illustrate the development of functionality and a user interface for the application. In addition, we discuss the advantages and limitations of the combined anatomic-vascular atlas.
|
Materials and Methods
|
|---|
The anatomic atlas was derived from print material, whereas the vascular atlas was constructed from angiographic data. The atlas materials were processed, and a user-friendly atlas-based application was developed.
Materials
Anatomic Atlas.—
The print Talairach-Tournoux atlas (26) containing gross anatomy on two-dimensional atlas plates was used to build the anatomic atlas. The print atlas was digitized, processed, fully color-coded, labeled, enhanced, and extended, and an anatomic index listing the anatomic structures was created. The development of the atlas has been addressed in more detail elsewhere (7,27). For the present work, a 3D version of the electronic Talairach-Tournoux brain atlas was used.
Vascular Atlas.—
A time-of-flight acquisition was performed on a Siemens Allegra 3T MR imager with a birdcage head coil and Syngo MR imaging software (Siemens, Milwaukee, Wis). The whole brain from the vertex to the bifurcation of the carotid arteries was scanned in five slabs. For each slab, 24 sections with a 1-mm section thickness and 0.42 x 0.42-mm2 pixel spacing were acquired (echo time msec/repetition time msec = 5/45, 18° flip angle). The data were processed as explained in the following paragraphs, and a 3D vascular model was constructed.
Methods
The application was developed with the following steps: (a) construction of 3D anatomic models, (b) construction of 3D vascular models, (c) spatial coregistration of the anatomic and vascular models, and (d) development of an atlas-based application with a suitable functionality and user interface.
A 3D anatomic atlas was derived from the Cerefy Electronic Brain Atlas Library (6), which contains the electronic version of the enhanced Talairach-Tournoux brain atlas (7,27). To build a 3D polygonal model of an individual structure, two-dimensional coronal images of the structure were smoothed out in 3D with Gaussian smoothing and then thresholded, and a 3D isosurface was extracted. A surface mesh representation was converted to the Macromedia Director MX representation (Macromedia, San Francisco, Calif).
The 3D vascular models were derived from angiographic data with the following steps: (a) vessel segmentation, (b) extraction of the centerline and radius, (c) centerline editing, (d) centerline smoothing, (e) radius processing, (f) modeling of vascular segments and bifurcations, and (g) labeling of vascular segments.
The blood vessels were segmented from the MR angiographic volumetric data set in three steps: image enhancement, adaptive thresholding, and editing. First, the volume was filtered based on fuzzy logic (28). Bright vascular candidates were enhanced by applying a nonlinear transform (29). The images were then converted into white-black representation (binarized) with adaptive thresholding. Finally, manual editing was performed to enhance the segmented vessels (Fig 1).
The centerline and radius were calculated based on a distance transform (30). The skeletonization process yielded a centerline one voxel in width.
Because of noise, inhomogeneity, and artifacts introduced by skeletonization, the centerline may contain spurious or disconnected branches. Manual editing was performed to remove very short vascular branches (less than three voxels long) having no "children" and to connect branches that were disconnected due to a low MR imaging signal intensity.
The centerline curve should have at least its first derivative continuous. However, the curve formed by connecting a sequence of 3D centerline points is typically not sufficiently smooth because of the limited accuracy of skeletonization, noise, and image inhomogeneity. Connecting the centerline points with cardinal splines results in a convoluted curve. B-splines produce intuitively unpredictable results due to variation in the position of the centerline points, making it difficult to adjust the shape of the resulting curve to its anatomic position. To smooth the centerline, we used a sliding average filter approach (31). Depending on the quality of the initial data, the smoothing can be repeated several times. After the first, second, and third use of the filter, the shift of the centerline is less than 0.5 mm, 0.81 mm, and 1.04 mm, respectively. In our application, a visually smooth model was obtained after three smoothing iterations.
After skeletonization, the radius along the centerline is scattered due to noise and the partial volume averaging effect (Fig 2a). It is known from anatomic studies that vessel radius decreases gradually (about 5%–10%) as one moves in the direction of blood flow. Hence, to perform radius smoothing, the direction of blood flow was first determined, followed by the removal of radius outliers. The variability of the radius along a blood vessel was plotted on a graph that gives the vessel radius as a function of distance along the centerline between consecutive bifurcations (ie, for a vessel segment) (Fig 2a). By studying several such graphs, we observed that the radius consistently changes slowly in the middle part of the vessel segment and more rapidly (over 20%) at the ends (ie, at the bifurcations [Fig 2a]). We also created a histogram illustrating the distribution of radius measurements (Fig 2b). The radius with the maximum frequency of occurrence represents the average radius of the middle part of the vessel segment. The radius of a short vessel (less than six voxels long) is defined as the average of its radii. For a vessel longer than six voxels, an adaptive window was applied to search for the starting and ending branching points. After the starting, middle, and ending parts of the vessel segment had been defined, linear regression was applied to the middle part. The outliers in this part were ignored during regression. After regression, these radii were linearly interpolated. Figure 2a shows the smoothed radius plot, and Figure 2c shows its corresponding histogram. The blood flow information and regression were used to keep the radii consistent (ie, nonincreasing) as one moved along the vessel in the direction of blood flow.
View larger version (16K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2a. Radius processing. (a) Graph illustrates the distribution of radial measurements along a sample vessel segment before and after smoothing. (b, c) Histograms show the distribution of radial measurements before (b) and after (c) smoothing.
|
|
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2b. Radius processing. (a) Graph illustrates the distribution of radial measurements along a sample vessel segment before and after smoothing. (b, c) Histograms show the distribution of radial measurements before (b) and after (c) smoothing.
|
|
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2c. Radius processing. (a) Graph illustrates the distribution of radial measurements along a sample vessel segment before and after smoothing. (b, c) Histograms show the distribution of radial measurements before (b) and after (c) smoothing.
|
|
Two different geometries, tubular and bifurcation, were distinguished in vascular modeling. The middle part of the vessel segment was modeled as a tubular object. The bifurcation, formed by three parts (one "parent" and its two children), was too complicated to be simplified as three crossing tubes. B-subdivision was applied to approximate the shape of the bifurcation. For the tubular part, the subsequent points along the centerline were connected with Kochanek–Bartels cubic splines (32). The tangential, normal, and binormal vectors of the tubular model were computed along this spline. The model for the tubular part was constructed as follows:
1. A circle was defined in the plane orthogonal to the tangent at a given centerpoint, with the radius corresponding to that of the vessel at this point.
2. The circumference was divided into 16 parts (this is adequate for good visualization, yields a smooth representation of the surface, and is computationally efficient).
3. These circumference points were connected with the points on the previously defined circle.
This tubular modeling requires a relatively small number of polygons and allows efficient rendering. To approximate bifurcation, B-subdivision is applied. The method described by Catmull and Clark (33) was used for bicubic uniform B-spline surface refinement. This procedure requires a comparatively large number of patches because each initial quadrangle patch is subdivided into 16 patches after two steps, but only for a small bifurcation region. We use quadrangles to represent the bifurcation surface and perform two steps of B-subdivision for surface refinement (Fig 3).
The vascular index was constructed on the basis of the terminology used in Terminologia Anatomica (34). A dedicated environment was developed for interactive labeling of vascular segments. All modeled vascular segments were subsequently labeled with their names.
All 3D models of subcortical structures (considered as a single compound object) were translated, rotated, and scaled interactively to fit into the vascular models.
|
Results
|
|---|
With use of Macromedia Director MX as an authoring tool, a user-friendly application was developed for exploration of the combined 3D anatomic-cerebrovascular atlas. Functions for interactive manipulation and querying of atlas structures were also developed. Interaction with the 3D models was implemented with built-in functions in Lingo (the programming language for Macromedia Director) that allow the user to rotate the model as well as move the model in or out, left or right, and up or down. Labeling of 3D anatomic structures and vascular segments was achieved by issuing a Lingo command returning the closest model along a normal line issued at the pointed location.
The user interface of the developed application contains the atlas viewing area, anatomic and vascular indexes, an atlas navigation panel, and a control panel (Fig 4). The atlas viewing area and both indexes are active, allowing the user to manipulate the atlas and its components.
The user has flexibility in making a selection from the atlas navigation panel. For instance, he or she can select all anatomic structures (Fig 5a) or all vascular segments (Fig 5b). In addition, the user can interactively label anatomic structures and vascular segments in the atlas viewing area (Fig 5). The user can also point to a structure or segment with the cursor to read its name or "click and drag" to draw a line with the name of the structure at its end. Furthermore, the user can freely label a 3D image and save it to an external file or select individual anatomic structures and vascular segments directly from the anatomic and vascular indexes (Fig 6).
|
Discussion
|
|---|
The vasculature of the human brain, including cerebrovascular variations, has been well studied radiologically (35–37). However, a digital cerebrovascular atlas that is widely accepted by the radiology community does not yet exist. We have been working toward achieving this goal. Our atlas-based application provides a user-friendly way to interactively explore the anatomy and vasculature of the brain. The user may choose the anatomic atlas, the vascular atlas, or the combined atlas for browsing. The use of polygonal models (with varying degrees of light and shade) allows smooth navigation by providing continuous rotate, pan, and zoom functions as well as an interactive labeling feature. Multiple labels can be placed on the 3D models in the atlas viewing area. Moreover, any anatomic structure or vascular segment can be selected from the anatomic or vascular index for exploration. The arranged image can be saved to an external file or imported into user applications, which is useful for medical students and for educators in the preparation of teaching materials.
The main advantage of the atlas is that it allows navigation the brain anatomy and vasculature as they relate to each other. For adequate learning, it is essential that the cerebral vessels be studied in relation to the brain parenchyma. Doing so not only provides a more detailed and realistic conception of the cerebral structures, but also gives the user a sense of having had a more complete learning experience. With the relationships between vessels and cerebral structures—especially the deep structures around the circle of Willis (Fig 7)—displayed in greater detail, users can gain a greater insight into relational neuroanatomy, which is essential for neurosurgical interventions. The main concern of neurosurgeons is awareness of arterial structures in the surgical field so as to avoid unnecessary bleeding. Our combined anatomic-vascular atlas allows neurosurgeons to gain a more complete understanding of the cerebral anatomic structures and to better appreciate the vasculature in relation to these structures.
The manipulation of the complete set of 3D models is quite smooth on a Pentium 4 2.4-GHz central processing unit (Intel, Santa Clara, Calif). However, Lingo is generally slower than applications based on the Java3D (Sun Microsystems, Santa Clara, Calif) or OpenGL (SGI, Mountain View, Calif) library.
The application has several limitations in terms of functionality and models. The labels placed in the atlas viewing area cannot yet be panned and rotated. Selection of arbitrary multiple structures or vascular segments is not feasible; only single structures can be selected. The appearance of structures is predefined, and the user has no interactive control over their lighting and shading. The 3D cortical areas are not included in the anatomic atlas, mainly due to inconsistency of the Talairach-Tournoux atlas in 3D. The content of the vascular atlas is still limited: Many arteries are missing due to either MR angiographic acquisition or insufficient image segmentation, and the veins are not yet available.
The majority of these limitations will soon be eliminated. New MR arteriographic and MR venographic data sets were acquired, and about 70 venous segments have recently been included in the application (38). A powerful tubular editor is being developed for 3D editing of segmented and modeled vascular structures, which will allow the incorporation of small vessels into the vascular atlas. A more advanced atlas-based application with an enhanced user interface is under development.
|
Summary
|
|---|
The combined anatomic-vascular 3D atlas described in this article enables the user to examine 3D anatomic structures and 3D cerebral vasculature and gain an understanding of their interrelationships. It is a user-friendly neuroeducational tool that is useful for medical students and neuroscience researchers as well as for educators who are preparing teaching materials.
|
TAKE-HOME POINTS
|
|---|
MR angiographic data and print information are highly preprocessed to create an interactive 3D combined atlas of brain anatomy and cerebral vasculature.
An efficient modeling technique and a user-friendly application make real-time manipulation of all the cerebral models feasible.
Two-way image-index labeling is provided, enabling the user to interactively name the cerebral structures on the 3D image or select them from the anatomic and vascular indexes. The structures of interest can be labeled in 3D and the image saved to an external file.
The user can select individual subcortical structures or vascular segments from the indexes for exploration.
|
Footnotes
|
|---|
Abbreviation: 3D = three-dimensional
|
References
|
|---|
- Sundsten JW, Brinkley JF, Eno K, Prothero J, Williams R. The digital anatomist: interactive brain atlas—CD ROM for Windows Seattle, Wash: University of Washington School of Medicine, 1996.
- Hoehne KH. VOXEL-MAN 3-D Navigator: brain and skull 2nd ed. Heidelberg, Germany: Springer-Verlag, 1997.
- Nowinski WL, Thirunavuukarasuu A, Kennedy DN. Brain atlas for functional imaging: clinical and research applications New York, NY: Thieme, 2000.
- Nowinski WL, Thirunavuukarasuu A, Bryan RN. The Cerefy atlas of brain anatomy: an interactive reference tool for students, teachers, and researchers New York, NY: Thieme, 2002.
- Nowinski WL, Thirunavuukarasuu A, Benabid AL. The Cerefy clinical brain atlas: extended edition with preoperative planning and intraoperative support New York, NY: Thieme, 2004.
- Cerefy atlas-based solutions. Available at: www.cerefy.com.
- Nowinski WL, Fang A, Nguyen BT, et al. Multiple brain atlas database and atlas-based neuroimaging system. Comput Aided Surg 1997; 2:42-66.[CrossRef][Medline]
- Nowinski WL. Computerized brain atlases for surgery of movement disorders. Semin Neurosurg 2001; 12:183-194.[CrossRef]
- Nowinski WL, Belov D, Benabid AL. An algorithm for rapid calculation of a probabilistic functional atlas of subcortical structures from electrophysiological data collected during functional neurosurgery procedures. Neuroimage 2003; 18:143-155.[CrossRef][Medline]
- Mazziotta J. The International Consortium for Brain Mapping: a probabilistic atlas and reference system for the human brain. In: Toga AW, Mazziotta JC, eds. Brain mapping: the methods. 2nd ed. Amsterdam, the Netherlands: Academic Press, 2002; 727-755.
- Thompson P, Mega MS, Toga AW. Subpopulation brain atlases. In: Toga AW, Mazziotta JC, eds. Brain mapping: the methods. 2nd ed. Amsterdam, the Netherlands: Academic Press, 2002; 757-796.
- Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J. Digital brain atlases. Trends Neurosci 1995; 18:210-211.[CrossRef][Medline]
- Mazziotta JC, Toga AW, Evans A, Fox P, Lancaster J. Probabilistic atlas of the human brain: theory and rationale for its development. Neuroimage 1995; 2:89-101.[CrossRef][Medline]
- Nowinski WL. Model enhanced neuroimaging: clinical, research, and educational applications. Yearbook of medical informatics. Stuttgart, Germany: Schattauer, 2002; 132-144.
- Nowinski WL, Yang GL, Yeo TT. Computer-aided stereotactic functional neurosurgery enhanced by the use of the multiple brain atlas database. IEEE Trans Med Imaging 2000; 19:62-69.[CrossRef][Medline]
- Nowinski WL. Anatomical targeting in functional neurosurgery by the simultaneous use of multiple Schaltenbrand-Wahren brain atlas microseries. Stereotact Funct Neurosurg 1998; 71:103-116.[CrossRef][Medline]
- Nowinski WL, Belov D. The Cerefy neuroradiology atlas: a Talairach-Tournoux atlas-based tool for analysis of neuroimages. Neuroimage 2003; 20:50-57.[CrossRef][Medline]
- Nowinski WL, Thirunavuukarasuu A. Electronic atlases show value in brain studies. Diagn Imaging Asia Pacific 2001; 2:35-39.
- Nowinski WL, Thirunavuukarasuu A. A locus-driven mechanism for rapid and automated atlas-assisted analysis of functional images by using the Brain Atlas for Functional Imaging. Neurosurg Focus 2003; 15:E3.
- Nowinski WL, Thirunavuukarasuu A. Atlas-assisted localization analysis of functional images. Med Image Anal 2001; 5:207-220.[CrossRef][Medline]
- Nowinski WL, Benabid AL. New directions in atlas-assisted stereotactic functional neurosurgery. In: Germano IM, eds. Advanced techniques in image-guided brain and spine surgery. New York, NY: Thieme, 2002; 162-174.
- Nowinski WL, Belov D, Benabid AL. A community-centric Internet portal for stereotactic and functional neurosurgery with a probabilistic functional atlas. Stereotact Funct Neurosurg 2002; 79:1-12.[CrossRef][Medline]
- Berkovitz B, Kirsch C, Moxham B, Alusi G, Cheeseman T. Interactive head & neck: CD-ROM PC and Mac compatible London, England: Primal Pictures, 2003.
- Kretschmann HJ, Weinrich W. Neurofunctional systems 3D reconstructions with correlated neuroimaging. CD-ROM for Windows 95, 98, and NT. Stuttgart, Germany: Thieme, 1999.
- Microvascular atlas of the head and neck: CD-ROM for Macintosh and Windows. Bayer 1996.
- Talairach J, Tournoux P. Co-planar stereotactic atlas of the human brain Stuttgart, Germany: Thieme, 1988.
- Nowinski WL. Electronic brain atlases: features and applications. In: Caramella D, Bartolozzi C, eds. 3D image processing: techniques and clinical applications—Medical Radiology series. Heidelberg, Germany: Springer-Verlag, 2002; 79-93.
- Choi YS, Krishnapuram R. A robust approach to image enhancement based on fuzzy logic. IEEE Trans Image Proc 1997; 6:808-825.[Medline]
- Hu QM, Hoad C, Sati M, Martel A, Nolte LP. Segmentation of spine MRI based on transforms and morphology. Proceedings of the 10th International Conference on Biomedical Engineering, Singapore 2000; 481-482.
- Hu QM. Two and three dimensional skeletonization: PCT patent application. International Publication Number WO 02/058008 2002.
- Yi D, Hayward V. Skeletonization of volumetric angiograms for display. Comput Methods Biomech Biomed Engin 2002; 5:329-341.[CrossRef][Medline]
- Kochanek DHU, Bartels RH. Interpolating splines with local tension, continuity, and bias control. Proc ACM SIGGRAPH 1984; 18:33-41.
- Catmull E, Clark J. Recursively generated B-spline surfaces on arbitrary topological meshes. Comput Aided Design 1978; 10:350-355.[CrossRef]
- Terminologia anatomica: international anatomical terminology. Stuttgart, Germany: Thieme 1998.
- Salamon G, Huang YP. Radiologic anatomy of the brain Heidelberg, Germany: Springer-Verlag, 1976.
- Krayenbuehl H, Yasargil MG, Huber P. Cerebral angiography 2nd ed. Stuttgart, Germany: Thieme, 1982.
- Grand W, Hopkins LN. Vasculature of the brain and cranial base: variations in clinical anatomy New York, NY: Thieme, 1999.
- Nowinski WL, Thirunavuukarasuu A, Ma X, et al. Anatomical, functional and vascular brain atlases as well as user-friendly tools for medical students, educators, clinicians and researchers. Final Programme, European Congress of Radiology 2004, The Matrix exhibition, 5–9 March 2004, Vienna, Austria (abstr). Eur Radiol 2004; 14(suppl 2):541.
Top
Abstract
Introduction
