- Email: [email protected]
Parametric display of myocardial function
Computerized Medical Imaging and Graphics PERGAMON
Computerized Medical Imaging and Graphics 25 (2001) 483±493
www.elsevier.com/locate/compmedimag
Parametric display of myocardial function Christian D. Eusemann a,b,*, Erik L. Ritman b, Matthias E. Bellemann b, Richard A. Robb a a
b
Mayo Foundation, 200 First Street SW, Rochester, MN 55905, USA University of Applied Sciences Jena, Carl-Zeiss Promenade 1b 07745 Jena, Germany Received 25 August 2000; accepted 9 February 2001
Abstract Quantitative assessment of regional heart motion has signi®cant potential to provide more speci®c diagnosis of cardiac disease and cardiac malfunction than currently possible. Local heart motion may be captured from various medical imaging scanners. In this study, 3-D reconstructions of pre-infarct and post-infarct hearts were obtained from the Dynamic Spatial Reconstructor [Ritman EL, Robb RA, Harris LD. Imaging physiological functions: experience with the DSR. Philadelphia: Praeger, 1985; Robb RA, Lent AH, Gilbert BK, Chu A. The dynamic spatial reconstructor: a computed tomography system for high-speed simultaneous scanning of multiple cross sections of the heart. J Med Syst 1980;4(2):253±88; Jorgensen SM, Whitlock SV, Thomas PJ, Roessler RW, Ritman EL. The dynamic spatial reconstructor: a high speed, stop action, 3-D, digital radiographic imager of moving internal organs and blood. Proceedings of SPIE, Ultrahigh- and High-speed Photography, Videography, Photonics, and Velocimetry 1990;1346:180±91.] (DSR). Using functional parametric mapping of disturbances in regional contractility and relaxation, regional myocardial motion during a cardiac cycle is color mapped onto a deformable heart model to facilitate appreciation of the structure-to-function relationships in the myocardium, such as occurs in regional patterns of akinesis or dyskinesis associated with myocardial ischemia or infarction resulting from coronary artery occlusion. q 2001 Elsevier Science Ltd. All rights reserved. Keywords: Myocardial dynamics; Functional mapping; Heart motion analysis
1. Introduction Cardiac disease is the most common cause of death. In 1996, cardiovascular disease claimed almost one million lives in the US alone [1], or 41% of all deaths for that year. The high mortality rate results from the number of patients affected by cardiac or coronary artery disease combined with inadequate treatment options. For accurate diagnosis and effective treatment of heart disease, information about the structure and regional mechanical behavior of the myocardium is important. Improvements in medical imaging technology provide rich opportunities for extraction of these clinically useful parameters. Previous studies have used 3-D data sets acquired from Electron Beam CT [2] MRI (especially MRI tagging [3±6]) Ultrasound [7,8] and DSR [9±13] for the non invasive assessment of dynamic heart motion. MRI tagging allows unambiguous tracking of myocardial segments. It is the method of choice for muscle segment motion tracking, and it provides useful assessment of myocardial motion, but is limited by the cost and relatively long scan times. Ultrasound is relatively inexpensive and fast (real-time), but until recently did not * Corresponding author. Tel.: 11-507-284-2997; fax: 11-507-284-1632. E-mail address: [email protected] (C.D. Eusemann).
provide full 3-D scans and suffers from relatively poor spatial resolution and speckle noise. The DSR was a unique, high temporal resolution, full 3-D scanner, a one-of-a-kind research system, but is no longer available. While current whole-body scanners can be used to provide 4D images of the in vivo heartsÐthe DSR scanner images are still the only detailed 3D images sequence obtained within one or two sequential cardiac cycles. This paper describes a method to measure and visualize myocardial wall structure and function using a fast, modalityindependent, algorithm applied to both normal and abnormal hearts in a controlled study. This new method can be applied to any 4-D data set (x, y, z, t) provided by rapid, repetitive, use of a fast 3-D scanning system. The algorithm is based on deformable surface reconstruction, involving creation of a polygonal surface mesh to represent the myocardial walls. The initial surface mesh represents the spatial position of individual surface elements of the cardiac walls. This triangular mesh deforms to ®t successive segmented cardiac volumes throughout the cardiac cycle. The method forces the mesh at each time point volume to have the same number of vertices and triangles. Therefore regional excursions and speeds of each vertex representing the same unit of myocardium can be calculated at each time-point interval. These parameters can be visualized
0895-6111/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved. PII: S 0895-611 1(01)00009-X
484
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
through color-mapped displays and animations of the dynamic myocardial model. The absolute value of regional maximum speed and excursion of myocardium during the cardiac cycle can be similarly computed and displayed. The maximum speed of left ventricular wall thickening is approximately 50 mm/s [14]. Animations provide visualization of local myocardial velocity changes and trajectories of myocardial surface points along speci®c pathways throughout the cardiac cycle. Using such parametric visualization methods, quantitative functional information is encoded on dynamic beating heart anatomy, which has potential for enhancing the diagnostic value of 4-D images of the heart, including improved diagnosis of myocardial abnormalities, such as regional ischemia, infarction, anomalous contrast ®lling defects and electrophysiologic disturbances in regional contraction patterns. To test and demonstrate the potential, data sets comprised of volume image reconstructions throughout a complete cardiac cycle of a canine heart before and after occlusion of an epicardial coronary artery was obtained with the DSR scanner. 2. Methods 2.1. Image acquisition For 4-D quantitative visualization of myocardial dynamics, 3-D reconstructions over time are necessary. Myocardial reconstructions were obtained with the DSR scanner at the Mayo Clinic. The DSR was a high-resolution three-dimensional imaging scanner, which operates on the principles of computerized tomography (CT). To achieve high speed, 14 equispaced X-ray tubes and imaging cameras, mounted on a gantry, rotates about the subject at 15 revolutions per minute. Accurate stop-action (0.011 s) 3-D images of moving organs can be achieved with the DSR at 15±30 volume images/s with better than 1 mm 3-D isotropic resolution. 3-D image data sets used for this study were obtained from DSR scans of the in situ canine heart of anesthetized dogs at 15 time points throughout one complete cardiac cycle. The heart rate was 60 beats/min. The ®rst data set was of a healthy, in vivo, canine heart. Each 1/15 s time point of this volume image included 110 images of contiguous cross section, each with 128 £ 128, 0.925 mm 3 voxels. The second data set was a canine heart scanned before and after an infarct was caused by acutely occluding the left anterior descending coronary artery (LAD). These data sets included 115 contiguous images with 0.75 mm 3 voxels. These data sets span one complete cardiac cycle, beginning and ending at end diastole. Contrast agents were injected intravenously to enhance the visibility of chambers and major vessels. 2.2. Image segmentation Segmentation of anatomic structures, which isolates speci®c objects in an image, is often required in image
analysis. Segmentation and display of the cardiac image volumes acquired in this study was performed using the `AnalyzeAVW' [15,16] software package developed in the Biomedical Imaging Resource at Mayo Clinic. Using the `Volume Render' tool, all adjacent slices obtained at one time point can be reconstructed and displayed as a single 3-D volume. Specifying an appropriate image grayscale threshold can adaptively segment the contrast ®lled chambers. But in order to divide the data into speci®c anatomic regions, such as the left ventricle (LV), atrium and aorta, various other segmentation tools are used, including manual tracing The disadvantages of manual segmentations are that it is labor intensive and is prone to subjective and/or fatigue errors. Using the `intelligent' manual `Trace Tool' and `Slice Edit' Tool in `AnalyzeAVW', regions of interest can be accurately traced with a computer mouse or trackball. The `Trace Tool' allows tracing, demarking and deleting speci®c regions in 3-D projections. Regions can be de®ned or erased all the way through the volume or by only one pixel or layer at a time. In `Slice Edit', areas of interest can be traced on single slices of the image volume in transverse, coronal or sagittal views. Both tools transform the data format from image ®les to object ®les for subsequent ef®cient manipulation and display. In manual segmentation, knowledge about the anatomic structures of the heart is essential. To determine the anatomical borders between the two chambers and between the LV and aorta, multiple anatomic landmarks are required. The 3-D image generally does not show the cardiac valves, which are the most useful structures to use as reference points for segmenting these structures. However, by comparing anatomy through time, the root of the aorta near the aortic valve is helpful. The papillary muscles were also used as reference points. Using these references the left ventricle, atrium and aorta were segmented throughout the cardiac cycle. As a ®nal step preparatory to applying the surface tiling algorithm on the segmented objects, a `Math 3-D Morphology tool' in `AnalyzeAVW' was used to ®ll any `holes' in the segmented objects, which would cause errors in the modeling process. Regions of lower contrast due to intravascular streaming of contrast agent may cause some `holes' during the thresholding step in the segmentation. 2.3. Surface reconstruction of the myocardium Surface reconstruction is the transformation from a segmented image volume to a geometric description of an object surface. Initially the segmented surface is identi®ed and the object transformed into a binary volume. To model the physical structure of the LV volumes, an adaptive deformable surface tiling program was used on the binary surface volumes [17]. The physical structure is represented as a set of nodes (polygon vertices) interconnected by adjustable springs (polygon edges) into a triangular mesh. The structure of the segmented volume at end diastole was used as the initial model. This ®rst triangular mesh deforms to ®t
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
485
Fig. 1. Polygonal mesh deformable model of left ventricle at end diastole (left), 0.27 s after end diastole (center) and at end systole (right) of a healthy canine heart beats at 60 beats per minute.
every volume throughout the cardiac cycle. The method forces the mesh for each time point volume to have the same number of vertices and triangles. Each surface vertex is assigned to the closest surface voxel of the subsequent time-point surface to drive the deformation to the surface. The displacement of a single vertex in the mesh can be used to describe the trajectory of motion of a speci®c myocardial surface point throughout the cardiac cycle. Fig. 1 shows the triangular mesh at the end diastolic, mid systolic and end systolic volumes of the normal heart data set. 2.4. Motion tracking, analysis and visualization The goal of this research is to track, analyze and visualize instantaneous regional myocardial motion throughout the cardiac cycle of both healthy and damaged hearts. To compute these parametric functions, SGI workstations are used, especially for modeling and visualization, which is compute intensive. Advanced software written in C and C11 is used on these workstations. 2.4.1. Tracking and analysis To measure motion, information about each vertex was assigned to a speci®c position in an array. The array was partitioned into ®fteen major columns representing the time points. Each column included three sub columns representing x, y, z coordinates of the vertex. The rows of the array contained the vertices. Each time point of the ®rst data set consisted of 11,660 vertices and 23,316 faces. The total dimension of the array was 11,660 £ 9, representing 104,940 total coordinates. To process the second data set, two arrays with even larger total dimensions were created. The non-infarct time points consisted of 12,663 vertices and 25,324 faces, and the infarct time points consisted of 13,191 vertices and 26,384 faces. The speci®c goal of the research was to compute and display the regional maximum speed and excursion of myocardium through the cardiac cycle, including the absolute value of these parameters. The myocardium shifts globally within the chest throughout
the cardiac cycle, and this has a confounding in¯uence on quantifying intrinsic cardiac motion. To minimize this problem, a static vertex describing the cardiac shift was determined. The epicardial apex, which is relatively stable throughout the heart cycle [18]. So the epicardial apex was chosen as a ®xed reference point. Using this anatomic ®ducial marker, the heart shift could be determined and used to compensate for myocardial motion. To calculate the speci®c motion of the LV, the deformation constraint was based on the surface reconstruction algorithm used. Since each vertex describes the pathway of a single surface point, computation of the dynamic properties of myocardial function is then straightforward. Regional excursion is de®ned as the vector length between any selected vertex in space and the initial position of the vertex. Regional velocity is the vector distance one vertex moves during successive time-points. To ®nd the time point of regional maximum velocity and excursion, the maximum value for each surface vertex was calculated throughout the cardiac cycle. The time point of this regional maximum was assigned a speci®c color value. Colors were also assigned to the maximum excursion and velocity values. 2.4.2. Visualization To visualize instantaneous regional myocardial motion, parametric displays were created with absolute value or time point equivalent colors. The color hues were selected to enable ready differentiation of adjacent regions exhibiting different parametric values. Although this selection is somewhat arbitrary, it facilitates ready identi®cation of the functional parameters mapped onto the myocardial surface. A color bar with numeric indexing is provided with each parametric display to quickly de®ne the color assigned to each mapped value (time, velocity, etc.). The color scheme also provides a global pattern for quickly recognizing regions of interest. Each dynamic property was divided into ®fteen reference colors, which in turn were assigned to each 1/15 s time point in the cardiac cycle. The excursion value colors were
486
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 2. Time point through cardiac cycle of regional maximum speed of normal myocardium.
assigned in 0.75 voxel (0.69 mm) per 67 ms scan interval increments, and the velocity value colors in 0.5 voxel (0.46 mm) increments. To display the dynamic properties, the assigned color values were mapped onto the polygonal surface mesh representing the myocardial wall. The color of each surface triangle was determined by interpolating the colors of the three connecting triangular vertices. Full surface images of these functional 3-D structure maps can be displayed as the six faces of a cube pulled apart to reveal all myocardial surfaces simultaneously from the six orthogonal directions. To create a dynamic picture of the parametric changes re¯ecting contraction and relaxation of the myocardial walls, an animation of the mapped model was created. This animation displays simultaneously the anatomic deformation and the instantaneous local myocardial speed changes, as well as the trajectory of the motion for individual segments of myocardium throughout the cardiac cycle. 3. Results Fig. 2 illustrates the function to structure mapping of regional myocardial speed during 15 different time points of the entire cardiac cycle for the healthy heart. The color distribution in each of the six views of the ventricular model corresponds to the maximum speed of regional myocardium over the ventricular wall. For example, in the `back' panel view of the cube, the red region indicates myocardium that
reached maximum speed at time point 10 during the period of rapid ®lling. The period of rapid ®lling of the ventricle starts immediately after end systole (time-point 8), when the ventricular pressure has fallen to low diastolic pressure values. Fig. 3 shows that this interval is shorter than 200 ms, which correlates to a slightly longer time period than the average ®rst third of diastole under the hemodynamic conditions present during the scan. For example, the number of model vertices reaching maximum velocity at 733 ms from end diastole, or just into the rapid ®lling phase, is about 3600. Fig. 4 illustrates that about 70% of the vertices of the myocardium reach their regional maximum velocity during this phase of rapid diastolic ®lling, and not in the phase of rapid systolic ejection. To illustrate the differences in myocardial motion caused by an infarct, Figs. 5±8 show regional myocardial excursions and velocity values before and after occlusion of an epicardial coronary artery. In Fig. 5, the regional maximum excursion of the myocardium through the cardiac cycle before occlusion is shown. Fig. 6 illustrates this same mapping after the occlusion. Both illustrations show that the major myocardial surface areas reach their maximum regional excursion near end systole (time-point 7). However, the infarcted heart myocardium shows signi®cant differences in maximum excursion at the apex of the heart. The bottom, left, back and right panel views reveal regions of early maximum excursion near the apex of the chamber. The number of surface points in this region at time intervals 1±4 is about three times higher than for the normal
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
487
Fig. 3. Graphical display of the number of model vertices per time-interval of regional maximum absolute speed.
myocardium. These regions of early maximum excursions also have a low absolute value, which is consistent with a reduction in ejection fraction. The spatial distribution of regional excursion data also suggests that regions of the myocardium, distant from the infarct, partially compensate for reduced LV function caused by localized damaged tissue. This delay is illustrated in all panels of the infarct heart. The front, right and top panels illustrate large regions of late maximum excursions, and the bottom, top, back and
the left panels show areas of maximum excursion as late as time-point 10. The basal region especially shows areas of late maximum excursion. This occurrence in the infarcted LV leads to a shorter diastolic period for the LV. Figs. 7 and 8 compare regional maximum speed values of myocardium through the cardiac cycle for the normal and the infarcted heart, respectively. The graph in Fig. 9 compares the number of vertices in the LV model against regional maximum speeds, for both normal and infarcted
Fig. 4. Graphical display of the percentage of all model vertices reaching maximum speed during rapid and during slow ®lling and ejection.
488
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 5. Time point through cardiac cycle map of regional maximum excursion of normal myocardium.
hearts. In region A (where speeds range from 0 to 33.8 mm/ s) in Fig. 9, the heart wall of the heart with infarct shows 100% increase in number of vertices, whereas in region B (speeds greater than 33.8 mm/s), the heart wall of the heart
with infarct shows 50% decrease in number of vertices. Comparing Figs. 7 and 8 reveals that the infarcted myocardium has a large area of contiguously reduced speed values (#33.8 mm/s), which is also depicted in the graph of Fig. 9.
Fig. 6. Time point through cardiac cycle map of regional maximum excursion of infarcted myocardium.
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
489
Fig. 7. Value of regional maximum speed of normal myocardium through cardiac cycleÐtime point interval 66.7 ms; spatial resolution 0.75 mm 3/voxel.
Fig. 8. Value of regional maximum speed of infarcted myocardium through cardiac cycleÐtime point interval 66.7 ms; spatial resolution 0.75 mm 3/ voxel.
490
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 9. Frequency histogram of the number of model vertices in the entire myocardium of the dog, before and after infarction.
In this instance the area comprised nearly two thirds of the total surface area of the heart with infarct. In contrast to the narrow spectrum of low velocities in the infarcted myocardium, in the normal myocardium a broad spectrum of values between 20.3 and 67.5 mm/s is observed. The reduced maximum speed values of the infarcted heart helps to explain the previous interpretations of Figs. 7 and 8, namely that the maximum excursion and the decrease of the maximum excursion value is extended over more of the myocardium in the infarcted heart. Fig. 10 illustrates yet another type of quantitative parametric display. Two model images are shown, with selected vector trajectories throughout the cardiac cycle during animation of the deformable models. The normal and infarcted hearts are captured after end systole in the period of rapid diastolic ®lling. White shades on the myocardial surface depict regions of little or no motion, and gray shades show relative velocity values of moving diastolic myocardiumÐthe deeper gray corresponding to higher speed and more rapid motion. The dark vertex trajectories illustrate the systolic (contraction) motion pathway of selected myocardial surface points, and the white vertex trajectories illustrate the diastolic (relaxation) motion pathway of the same myocardial surface points through one complete heart cycle. 4. Discussion When local heart wall motion is imaged with 3D imaging devices, functional parametric mapping, spatial patterns of local myocardial motion during a cardiac cycle can be color mapped into a deforming heart model to obtain a better understanding of the structure-to-function relationship in the myocardium. 3D reconstructions were obtained from the Dynamic Spatial Reconstructor at 67 ms intervals (15 time points)
through out one cardiac cycle. This short time interval allows only very small displacement of local myocardium between consecutive time points, which results in a distinct improvement of smoothness and accuracy of myocardial motion patterns over large displacements using time intervals of 201 ms (5 time points). Initial processing involved segmentation and tiling of the left ventricle. Deformable models were created for each time point of a single cardiac cycle. An initial triangular mesh gradually deforms to ®t every volume. Through this process regional excursions and speeds of each vertex representing a piece of myocardium can be calculated for each time-point interval throughout the entire cardiac cycle. Since the algorithm is indented to be modality independent, it should also be applicable to MR-Tagging data. Such data would allow an evaluation of the accuracy of the surface point-tracking algorithm. The algorithm is computationally ef®cient, producing each functional structure map throughout the complete cardiac cycle in less than two minutes. To animate the entire heartbeat, with encoded functional mappings and trajectory of the motion for individual segments, less than three minutes is required on a SGI workstation. Thus the algorithm processes data in a time frame short enough for immediate data processing, after patient data acquisition, for potential clinical application on a routine basis. The calculated results can be visualized through model animations and specially formatted static images. Regional maximum speed and excursion of myocardium through the cardiac cycle can be statically displayed through color mapping. In addition the value of regional maximum speed and excursion of myocardium during the cardiac cycle can be displayed through color mapping. Using animations, the local myocardial speed changes can be visualized through color change on the cardiac surface during the cardiac cycle. Moreover, the trajectory of
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
491
Fig. 10. Display of the speed map and pathway trajectories of selected voxels during myocardial contraction (dark) and relaxation (white) on normal (left) and infarcted (right) infarction myocardiumÐwhite shades on the myocardial surface depict regions of little or no motion, and gray shades show relative speed values of moving myocardiumÐthe deeper gray corresponding to higher speed and more rapid motion.
motion for individual segments of myocardium can be displayed. 5. Summary The ability to encode quantitative functional information on a dynamic beating heart enhances the diagnostic value of 4D images in the cardiac cycle. Information regarding myocardial wall mechanics conveyed by such parametric displays adds another dimension to the assessment of some cardiac diseases. The ability to measure and display the motion of the heart walls combined with local myocardial excursion and speed maps could improve diagnosis of myocardial abnormalities, including regional ischemia, infarcts, akinesis and dyskinesis, diastolic contrast ®lling defects and/or electrophysiologic disturbances in regional conduction and contraction patterns. Acknowledgements The author expresses thanks to Wei-te Lin for his help with the deformable surface reconstruction program, to Pat Lund for assistance with segmentation and Mike Wahl, MD and Jon J. Camp for expert advice and technical support.
References [1] American Heart Association. Heart and StrokeÐStatistical Update, 1999. [2] Feiring AJ, Rumberger JA, Reiter SJ, Collins SM, Skorton DJ, Rees M, Marcus ML. Sectional and segmental variability of left ventricular function: experimental and clinical studies using ultra fast computed tomography. Journal of the American College of Cardiology. 1988;12(2):415±25. [3] Prince JL, McVeigh ER. Motion estimation from tagged MR image sequences. IEEE Transactions on Medical Imaging 1992;11:238±49. [4] Park J, Metaxas D, Axel L. Volumetric deformable models with parameter functions: a new approach to the 3D motion analysis of the LV from MRI-SPAMM. Fifth International Conference on Computer Vission, 1995. p. 700±5. [5] McEachen JC, Meyer FG, Constable RT, Nehorai A, Duncan JS. A recursive ®lter for phase velocity assisted shape-based tracking of cardiac non-rigid motion. http://noodle.med.yale.edu/mceachen/iccv/ iccv.ps.gz [6] Croissile P, Moore CC, Judd RM, Lima JAC, Arai M, McVeigh ER, Becker LC, Zerhouni EA. Differentiation of viable and nonviable myocardium by the use of three-dimensional tagged MRI in 2-dayold reperfused infarcts. Circulation 1999;99:284±91. [7] Papademetris X, Sinusas AJ, Dione DP, Duncan JS. 3D cardiac deformation from ultrasound images. Proceedings of MICCAI 1999;1679:420±9. [8] Mor-Avi V, Vigon P, Koch R, Weinert L, Garcia MJ, Spencer KT, Lang RM. Segmental analysis of color kinesis images: new method for quanti®cation of the magnitude and timing of endocardial motion
492
[9] [10]
[11]
[12]
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493 during left ventricular systole and diastole. Circulation. 1997; 95(8):2082±97. Shi P, Robinson G, Duncan JS. Myocardial motion and function assessment using 4D images. Proceedings of IEEE Visualization in Biomedical Computing 1994 Proc SPIE 1994;2359:148±59. Shi P, Amini A, Robinson G, Sinusas A, Constable CT, Duncan JS. Shape-based 4D left ventricular myocardial function analysis. Shapebased 4d left ventricular myocardial function analysis. Proceedings of the IEEE Workshop on Biomedical Image Analysis, Seattle, WA. IEEE 1994;327:604±12. Friboulet D, Magnin IE, Mathieu C, Pommert A, Hoehne KH. Assessment and visualization of the curvature of the left ventricle from 3D medical images. Computerized Medical Imaging and Graphics 1993;4/5:257±62. Gorce JM, Friboulet D, Clarysse P, Magnin IE. Three-dimensional velocity ®eld estimation of moving cardiac walls. Computers in Cardiology 1994;0276-6547/94:489±92.
[13] Clarysse P, Jaouen O, Magnin IE, Morvan JM. 3D representation and deformation analysis of the heart walls from X-ray and MRI images. Computers in Cardiology 1994;0276-6547/94:657±66. [14] St John Sutton MG, Ritman EL. Effects of progressive reduction in coronary blood ¯ow on regional and global left ventricular contraction and relaxation during normal and increased afterload: a roentgen videometric study. Cardiovascular Research 1982;16(9):535±45. [15] Robb RA, Hanson DP. The ANALYZE software system for visualization and analysis in surgery simulation. In Computer Integrated Surgery. Cambridge, MA: MIT Press, 1995. [16] Robb RA. Three-dimensional biomedical imagingÐprinciples and practice. New York, NY: VCH Publishers, 1995. [17] Lin WT, Robb RA. Realistic visualization for surgery simulation using dynamic volume texture mapping and model deformation. Proceedings of SPIE, Medical Imaging 1999;3658:308±31. [18] Hoffman EA, Ritman EL. Invariant total heart volume in the intact thorax. Am J Physiol (Heart Circ Physiol 18) 1985;249:H883±90.
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493 Christian D. Eusemann received his Diplom Ingenieur Fachhochschule from the University of Applied Sciences of Jena, Germany in 1999. He is currently a Graduate Student working on his PhD in Biomedical Engineering at Mayo Graduate School. He was the recipient of a Carl Duisberg Society Research Fellowship Abroad for University of Applied Sciences Students in Germany and a Whitaker Foundation Student Travel Award for SPIE 2001 Image-Guided Procedures meeting. His research interest is in Cardiac Functional Imaging, and he has published four papers in this ®eld.
Matthias E. Bellemann studied Physics, Mathematics, and Computer Science at the University of Heidelberg, Germany. During his studies, he received a renowned scholarship from the German National Scholarship Foundation (Studienstiftung des deutschen Volkes). Having obtained his Diploma degree in Physics in 1989, he started his studies in Medicine at the Medical School of the University of Heidelberg. In 1992 he received his PhD degree in Biophysics jointly from the Max-Planck-Institute of Experimental Medicine and the University of Heidelberg. During his academic and professional education, Dr Bellemann joined the Max-Planck Institute of Nuclear Physics in Heidelberg, the European Molecular Biology Laboratory in Hamburg, and the German Cancer Research Center in Heidelberg. In 1997 he was appointed as a full Professor of Medical Physics at the University of Applied Sciences, Jena, Germany. He is currently Associate Dean for Research Affairs at the Department of Biomedical Engineering at the University of Applied Sciences. His research interests are focused on the development and application of advanced imaging techniques for quantitative mapping of functional biomedical data. He is a member of the International Society for Magnetic Resonance in Medicine, the German Roentgen Society, the German Society for Nuclear Medicine, and the German Society for Medical Physics. He is reviewer of several prestigious national and international journals. He has been and is principal investigator on several national research grants. He has authored or co-authored over 40 peer-reviewed journal articles in the ®eld of biomedical imaging. Dr Bellemann's research interests are in functional biomedical imaging techniques (MRI, PET, CT, and US), image analysis (registration, parameter mapping), signal processing (biostatistics, pharmacokinetic modeling), and application of these procedures in experimental and clinical trials. He has developed several speci®c clinical applications of these techniques, including carbon-13 MR spectroscopy for tumor characterization and monitoring, pharmacokinetic modeling of PET data for follow-up studies during therapy, 3D image registration in MR mammography, and electromagnetic ®eld calculation for imageguided transcranial magnetic stimulation. His research interests also include the development and evaluation of new inter-modality imaging techniques as well as the combined application of diagnostic and therapeutic procedures.
493
Erik L. Ritman is Professor of Physiology and Medicine and is the Ralph B. and Ruth K. Abrams Professor at Mayo Medical School. He graduated in Medicine at the University of Melbourne, Australia, in 1964 and obtained his PhD in Physiology in 1973 at the University of Minnesota. His main interest is development of imaging devices for quantitative evaluation of cardiovascular structure-to-function relationships. In 1977 he became the principal investigator of the National Institutes of Healthprogram project grant that funded the fabrication and evaluation of the DSR, a high speed 3D CT scanner for study of humans and large animals. He currently heads a multidisciplinary NIH research grant which supports development and applications of X-ray micro-CT methods for study of small rodents' organs and biopsies from humans and large animals.
Richard A. Robb received the BA degree in Mathematics in 1965, the MS degree in Computer Science in 1968, and the PhD degree in Computer Science and Biophysics in 1971, all from the University of Utah. He is currently the Scheller Professor in Medical Research and Professor of Biophysics and Professor of Computer Science in the Mayo Medical School and Mayo Graduate School. He is Associate Dean for Academic Affairs in the Mayo Graduate School. He is Director of the Biomedical Engineering Program and Director of the Mayo Biomedical Imaging Resource at Mayo Foundation/Clinic. He has been involved in the development and application of computer systems for processing, analysis, and display of biomedical image data for over 28 years. He is a member of the American Physiological Society, the Biomedical Engineering Society, the Institute of Electrical and Electronics Engineers, the Society of Photo-Optical Instrumentation Engineers, the American Association for the Advancement of Science, the Association for Computing Machinery, the National Computer Graphics Association and the International Society for Computer Assisted Surgery. He is editor, associate editor and on the editorial board of several prestigious international journals. He has been and is principal investigator on several NIH research grants and has over 300 publications in the ®eld of biomedical image processing, including ®ve books and 30 book chapters. He has patented several inventions related to display, manipulation and analysis of computer-generated medical images. He has directed development of comprehensive software packages, which provide advanced capabilities for multidimensional biomedical image visualization and analysis. These software packages are used in over 300 institutions around the world and have been licensed to several commercial companies. Dr Robb's research interests are in biomedical image visualization; image processing (segmentation, registration and classi®cation), image modeling; virtual reality; advanced software systems; and distributed high performance computing systems and networks for biomedical image analysis. He has developed several speci®c clinical applications of these techniques, including 3D image-guided neurosurgery for brain cancer and epilepsy, prostate cancer diagnosis and treatment, quanti®cation and treatment of coronary artery disease, catheter-based myocardial ablation, radiation therapy planning, craniofacial reconstructive surgery and computerized histological analysis. His interests also include design and evaluation of new-generation paradigms for biomedical visualization systems of the future, particularly for medical treatment planning, minimally invasive clinical procedures, computer assisted surgery, and medical education.
Computerized Medical Imaging and Graphics 25 (2001) 483±493
www.elsevier.com/locate/compmedimag
Parametric display of myocardial function Christian D. Eusemann a,b,*, Erik L. Ritman b, Matthias E. Bellemann b, Richard A. Robb a a
b
Mayo Foundation, 200 First Street SW, Rochester, MN 55905, USA University of Applied Sciences Jena, Carl-Zeiss Promenade 1b 07745 Jena, Germany Received 25 August 2000; accepted 9 February 2001
Abstract Quantitative assessment of regional heart motion has signi®cant potential to provide more speci®c diagnosis of cardiac disease and cardiac malfunction than currently possible. Local heart motion may be captured from various medical imaging scanners. In this study, 3-D reconstructions of pre-infarct and post-infarct hearts were obtained from the Dynamic Spatial Reconstructor [Ritman EL, Robb RA, Harris LD. Imaging physiological functions: experience with the DSR. Philadelphia: Praeger, 1985; Robb RA, Lent AH, Gilbert BK, Chu A. The dynamic spatial reconstructor: a computed tomography system for high-speed simultaneous scanning of multiple cross sections of the heart. J Med Syst 1980;4(2):253±88; Jorgensen SM, Whitlock SV, Thomas PJ, Roessler RW, Ritman EL. The dynamic spatial reconstructor: a high speed, stop action, 3-D, digital radiographic imager of moving internal organs and blood. Proceedings of SPIE, Ultrahigh- and High-speed Photography, Videography, Photonics, and Velocimetry 1990;1346:180±91.] (DSR). Using functional parametric mapping of disturbances in regional contractility and relaxation, regional myocardial motion during a cardiac cycle is color mapped onto a deformable heart model to facilitate appreciation of the structure-to-function relationships in the myocardium, such as occurs in regional patterns of akinesis or dyskinesis associated with myocardial ischemia or infarction resulting from coronary artery occlusion. q 2001 Elsevier Science Ltd. All rights reserved. Keywords: Myocardial dynamics; Functional mapping; Heart motion analysis
1. Introduction Cardiac disease is the most common cause of death. In 1996, cardiovascular disease claimed almost one million lives in the US alone [1], or 41% of all deaths for that year. The high mortality rate results from the number of patients affected by cardiac or coronary artery disease combined with inadequate treatment options. For accurate diagnosis and effective treatment of heart disease, information about the structure and regional mechanical behavior of the myocardium is important. Improvements in medical imaging technology provide rich opportunities for extraction of these clinically useful parameters. Previous studies have used 3-D data sets acquired from Electron Beam CT [2] MRI (especially MRI tagging [3±6]) Ultrasound [7,8] and DSR [9±13] for the non invasive assessment of dynamic heart motion. MRI tagging allows unambiguous tracking of myocardial segments. It is the method of choice for muscle segment motion tracking, and it provides useful assessment of myocardial motion, but is limited by the cost and relatively long scan times. Ultrasound is relatively inexpensive and fast (real-time), but until recently did not * Corresponding author. Tel.: 11-507-284-2997; fax: 11-507-284-1632. E-mail address: [email protected] (C.D. Eusemann).
provide full 3-D scans and suffers from relatively poor spatial resolution and speckle noise. The DSR was a unique, high temporal resolution, full 3-D scanner, a one-of-a-kind research system, but is no longer available. While current whole-body scanners can be used to provide 4D images of the in vivo heartsÐthe DSR scanner images are still the only detailed 3D images sequence obtained within one or two sequential cardiac cycles. This paper describes a method to measure and visualize myocardial wall structure and function using a fast, modalityindependent, algorithm applied to both normal and abnormal hearts in a controlled study. This new method can be applied to any 4-D data set (x, y, z, t) provided by rapid, repetitive, use of a fast 3-D scanning system. The algorithm is based on deformable surface reconstruction, involving creation of a polygonal surface mesh to represent the myocardial walls. The initial surface mesh represents the spatial position of individual surface elements of the cardiac walls. This triangular mesh deforms to ®t successive segmented cardiac volumes throughout the cardiac cycle. The method forces the mesh at each time point volume to have the same number of vertices and triangles. Therefore regional excursions and speeds of each vertex representing the same unit of myocardium can be calculated at each time-point interval. These parameters can be visualized
0895-6111/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved. PII: S 0895-611 1(01)00009-X
484
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
through color-mapped displays and animations of the dynamic myocardial model. The absolute value of regional maximum speed and excursion of myocardium during the cardiac cycle can be similarly computed and displayed. The maximum speed of left ventricular wall thickening is approximately 50 mm/s [14]. Animations provide visualization of local myocardial velocity changes and trajectories of myocardial surface points along speci®c pathways throughout the cardiac cycle. Using such parametric visualization methods, quantitative functional information is encoded on dynamic beating heart anatomy, which has potential for enhancing the diagnostic value of 4-D images of the heart, including improved diagnosis of myocardial abnormalities, such as regional ischemia, infarction, anomalous contrast ®lling defects and electrophysiologic disturbances in regional contraction patterns. To test and demonstrate the potential, data sets comprised of volume image reconstructions throughout a complete cardiac cycle of a canine heart before and after occlusion of an epicardial coronary artery was obtained with the DSR scanner. 2. Methods 2.1. Image acquisition For 4-D quantitative visualization of myocardial dynamics, 3-D reconstructions over time are necessary. Myocardial reconstructions were obtained with the DSR scanner at the Mayo Clinic. The DSR was a high-resolution three-dimensional imaging scanner, which operates on the principles of computerized tomography (CT). To achieve high speed, 14 equispaced X-ray tubes and imaging cameras, mounted on a gantry, rotates about the subject at 15 revolutions per minute. Accurate stop-action (0.011 s) 3-D images of moving organs can be achieved with the DSR at 15±30 volume images/s with better than 1 mm 3-D isotropic resolution. 3-D image data sets used for this study were obtained from DSR scans of the in situ canine heart of anesthetized dogs at 15 time points throughout one complete cardiac cycle. The heart rate was 60 beats/min. The ®rst data set was of a healthy, in vivo, canine heart. Each 1/15 s time point of this volume image included 110 images of contiguous cross section, each with 128 £ 128, 0.925 mm 3 voxels. The second data set was a canine heart scanned before and after an infarct was caused by acutely occluding the left anterior descending coronary artery (LAD). These data sets included 115 contiguous images with 0.75 mm 3 voxels. These data sets span one complete cardiac cycle, beginning and ending at end diastole. Contrast agents were injected intravenously to enhance the visibility of chambers and major vessels. 2.2. Image segmentation Segmentation of anatomic structures, which isolates speci®c objects in an image, is often required in image
analysis. Segmentation and display of the cardiac image volumes acquired in this study was performed using the `AnalyzeAVW' [15,16] software package developed in the Biomedical Imaging Resource at Mayo Clinic. Using the `Volume Render' tool, all adjacent slices obtained at one time point can be reconstructed and displayed as a single 3-D volume. Specifying an appropriate image grayscale threshold can adaptively segment the contrast ®lled chambers. But in order to divide the data into speci®c anatomic regions, such as the left ventricle (LV), atrium and aorta, various other segmentation tools are used, including manual tracing The disadvantages of manual segmentations are that it is labor intensive and is prone to subjective and/or fatigue errors. Using the `intelligent' manual `Trace Tool' and `Slice Edit' Tool in `AnalyzeAVW', regions of interest can be accurately traced with a computer mouse or trackball. The `Trace Tool' allows tracing, demarking and deleting speci®c regions in 3-D projections. Regions can be de®ned or erased all the way through the volume or by only one pixel or layer at a time. In `Slice Edit', areas of interest can be traced on single slices of the image volume in transverse, coronal or sagittal views. Both tools transform the data format from image ®les to object ®les for subsequent ef®cient manipulation and display. In manual segmentation, knowledge about the anatomic structures of the heart is essential. To determine the anatomical borders between the two chambers and between the LV and aorta, multiple anatomic landmarks are required. The 3-D image generally does not show the cardiac valves, which are the most useful structures to use as reference points for segmenting these structures. However, by comparing anatomy through time, the root of the aorta near the aortic valve is helpful. The papillary muscles were also used as reference points. Using these references the left ventricle, atrium and aorta were segmented throughout the cardiac cycle. As a ®nal step preparatory to applying the surface tiling algorithm on the segmented objects, a `Math 3-D Morphology tool' in `AnalyzeAVW' was used to ®ll any `holes' in the segmented objects, which would cause errors in the modeling process. Regions of lower contrast due to intravascular streaming of contrast agent may cause some `holes' during the thresholding step in the segmentation. 2.3. Surface reconstruction of the myocardium Surface reconstruction is the transformation from a segmented image volume to a geometric description of an object surface. Initially the segmented surface is identi®ed and the object transformed into a binary volume. To model the physical structure of the LV volumes, an adaptive deformable surface tiling program was used on the binary surface volumes [17]. The physical structure is represented as a set of nodes (polygon vertices) interconnected by adjustable springs (polygon edges) into a triangular mesh. The structure of the segmented volume at end diastole was used as the initial model. This ®rst triangular mesh deforms to ®t
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
485
Fig. 1. Polygonal mesh deformable model of left ventricle at end diastole (left), 0.27 s after end diastole (center) and at end systole (right) of a healthy canine heart beats at 60 beats per minute.
every volume throughout the cardiac cycle. The method forces the mesh for each time point volume to have the same number of vertices and triangles. Each surface vertex is assigned to the closest surface voxel of the subsequent time-point surface to drive the deformation to the surface. The displacement of a single vertex in the mesh can be used to describe the trajectory of motion of a speci®c myocardial surface point throughout the cardiac cycle. Fig. 1 shows the triangular mesh at the end diastolic, mid systolic and end systolic volumes of the normal heart data set. 2.4. Motion tracking, analysis and visualization The goal of this research is to track, analyze and visualize instantaneous regional myocardial motion throughout the cardiac cycle of both healthy and damaged hearts. To compute these parametric functions, SGI workstations are used, especially for modeling and visualization, which is compute intensive. Advanced software written in C and C11 is used on these workstations. 2.4.1. Tracking and analysis To measure motion, information about each vertex was assigned to a speci®c position in an array. The array was partitioned into ®fteen major columns representing the time points. Each column included three sub columns representing x, y, z coordinates of the vertex. The rows of the array contained the vertices. Each time point of the ®rst data set consisted of 11,660 vertices and 23,316 faces. The total dimension of the array was 11,660 £ 9, representing 104,940 total coordinates. To process the second data set, two arrays with even larger total dimensions were created. The non-infarct time points consisted of 12,663 vertices and 25,324 faces, and the infarct time points consisted of 13,191 vertices and 26,384 faces. The speci®c goal of the research was to compute and display the regional maximum speed and excursion of myocardium through the cardiac cycle, including the absolute value of these parameters. The myocardium shifts globally within the chest throughout
the cardiac cycle, and this has a confounding in¯uence on quantifying intrinsic cardiac motion. To minimize this problem, a static vertex describing the cardiac shift was determined. The epicardial apex, which is relatively stable throughout the heart cycle [18]. So the epicardial apex was chosen as a ®xed reference point. Using this anatomic ®ducial marker, the heart shift could be determined and used to compensate for myocardial motion. To calculate the speci®c motion of the LV, the deformation constraint was based on the surface reconstruction algorithm used. Since each vertex describes the pathway of a single surface point, computation of the dynamic properties of myocardial function is then straightforward. Regional excursion is de®ned as the vector length between any selected vertex in space and the initial position of the vertex. Regional velocity is the vector distance one vertex moves during successive time-points. To ®nd the time point of regional maximum velocity and excursion, the maximum value for each surface vertex was calculated throughout the cardiac cycle. The time point of this regional maximum was assigned a speci®c color value. Colors were also assigned to the maximum excursion and velocity values. 2.4.2. Visualization To visualize instantaneous regional myocardial motion, parametric displays were created with absolute value or time point equivalent colors. The color hues were selected to enable ready differentiation of adjacent regions exhibiting different parametric values. Although this selection is somewhat arbitrary, it facilitates ready identi®cation of the functional parameters mapped onto the myocardial surface. A color bar with numeric indexing is provided with each parametric display to quickly de®ne the color assigned to each mapped value (time, velocity, etc.). The color scheme also provides a global pattern for quickly recognizing regions of interest. Each dynamic property was divided into ®fteen reference colors, which in turn were assigned to each 1/15 s time point in the cardiac cycle. The excursion value colors were
486
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 2. Time point through cardiac cycle of regional maximum speed of normal myocardium.
assigned in 0.75 voxel (0.69 mm) per 67 ms scan interval increments, and the velocity value colors in 0.5 voxel (0.46 mm) increments. To display the dynamic properties, the assigned color values were mapped onto the polygonal surface mesh representing the myocardial wall. The color of each surface triangle was determined by interpolating the colors of the three connecting triangular vertices. Full surface images of these functional 3-D structure maps can be displayed as the six faces of a cube pulled apart to reveal all myocardial surfaces simultaneously from the six orthogonal directions. To create a dynamic picture of the parametric changes re¯ecting contraction and relaxation of the myocardial walls, an animation of the mapped model was created. This animation displays simultaneously the anatomic deformation and the instantaneous local myocardial speed changes, as well as the trajectory of the motion for individual segments of myocardium throughout the cardiac cycle. 3. Results Fig. 2 illustrates the function to structure mapping of regional myocardial speed during 15 different time points of the entire cardiac cycle for the healthy heart. The color distribution in each of the six views of the ventricular model corresponds to the maximum speed of regional myocardium over the ventricular wall. For example, in the `back' panel view of the cube, the red region indicates myocardium that
reached maximum speed at time point 10 during the period of rapid ®lling. The period of rapid ®lling of the ventricle starts immediately after end systole (time-point 8), when the ventricular pressure has fallen to low diastolic pressure values. Fig. 3 shows that this interval is shorter than 200 ms, which correlates to a slightly longer time period than the average ®rst third of diastole under the hemodynamic conditions present during the scan. For example, the number of model vertices reaching maximum velocity at 733 ms from end diastole, or just into the rapid ®lling phase, is about 3600. Fig. 4 illustrates that about 70% of the vertices of the myocardium reach their regional maximum velocity during this phase of rapid diastolic ®lling, and not in the phase of rapid systolic ejection. To illustrate the differences in myocardial motion caused by an infarct, Figs. 5±8 show regional myocardial excursions and velocity values before and after occlusion of an epicardial coronary artery. In Fig. 5, the regional maximum excursion of the myocardium through the cardiac cycle before occlusion is shown. Fig. 6 illustrates this same mapping after the occlusion. Both illustrations show that the major myocardial surface areas reach their maximum regional excursion near end systole (time-point 7). However, the infarcted heart myocardium shows signi®cant differences in maximum excursion at the apex of the heart. The bottom, left, back and right panel views reveal regions of early maximum excursion near the apex of the chamber. The number of surface points in this region at time intervals 1±4 is about three times higher than for the normal
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
487
Fig. 3. Graphical display of the number of model vertices per time-interval of regional maximum absolute speed.
myocardium. These regions of early maximum excursions also have a low absolute value, which is consistent with a reduction in ejection fraction. The spatial distribution of regional excursion data also suggests that regions of the myocardium, distant from the infarct, partially compensate for reduced LV function caused by localized damaged tissue. This delay is illustrated in all panels of the infarct heart. The front, right and top panels illustrate large regions of late maximum excursions, and the bottom, top, back and
the left panels show areas of maximum excursion as late as time-point 10. The basal region especially shows areas of late maximum excursion. This occurrence in the infarcted LV leads to a shorter diastolic period for the LV. Figs. 7 and 8 compare regional maximum speed values of myocardium through the cardiac cycle for the normal and the infarcted heart, respectively. The graph in Fig. 9 compares the number of vertices in the LV model against regional maximum speeds, for both normal and infarcted
Fig. 4. Graphical display of the percentage of all model vertices reaching maximum speed during rapid and during slow ®lling and ejection.
488
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 5. Time point through cardiac cycle map of regional maximum excursion of normal myocardium.
hearts. In region A (where speeds range from 0 to 33.8 mm/ s) in Fig. 9, the heart wall of the heart with infarct shows 100% increase in number of vertices, whereas in region B (speeds greater than 33.8 mm/s), the heart wall of the heart
with infarct shows 50% decrease in number of vertices. Comparing Figs. 7 and 8 reveals that the infarcted myocardium has a large area of contiguously reduced speed values (#33.8 mm/s), which is also depicted in the graph of Fig. 9.
Fig. 6. Time point through cardiac cycle map of regional maximum excursion of infarcted myocardium.
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
489
Fig. 7. Value of regional maximum speed of normal myocardium through cardiac cycleÐtime point interval 66.7 ms; spatial resolution 0.75 mm 3/voxel.
Fig. 8. Value of regional maximum speed of infarcted myocardium through cardiac cycleÐtime point interval 66.7 ms; spatial resolution 0.75 mm 3/ voxel.
490
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
Fig. 9. Frequency histogram of the number of model vertices in the entire myocardium of the dog, before and after infarction.
In this instance the area comprised nearly two thirds of the total surface area of the heart with infarct. In contrast to the narrow spectrum of low velocities in the infarcted myocardium, in the normal myocardium a broad spectrum of values between 20.3 and 67.5 mm/s is observed. The reduced maximum speed values of the infarcted heart helps to explain the previous interpretations of Figs. 7 and 8, namely that the maximum excursion and the decrease of the maximum excursion value is extended over more of the myocardium in the infarcted heart. Fig. 10 illustrates yet another type of quantitative parametric display. Two model images are shown, with selected vector trajectories throughout the cardiac cycle during animation of the deformable models. The normal and infarcted hearts are captured after end systole in the period of rapid diastolic ®lling. White shades on the myocardial surface depict regions of little or no motion, and gray shades show relative velocity values of moving diastolic myocardiumÐthe deeper gray corresponding to higher speed and more rapid motion. The dark vertex trajectories illustrate the systolic (contraction) motion pathway of selected myocardial surface points, and the white vertex trajectories illustrate the diastolic (relaxation) motion pathway of the same myocardial surface points through one complete heart cycle. 4. Discussion When local heart wall motion is imaged with 3D imaging devices, functional parametric mapping, spatial patterns of local myocardial motion during a cardiac cycle can be color mapped into a deforming heart model to obtain a better understanding of the structure-to-function relationship in the myocardium. 3D reconstructions were obtained from the Dynamic Spatial Reconstructor at 67 ms intervals (15 time points)
through out one cardiac cycle. This short time interval allows only very small displacement of local myocardium between consecutive time points, which results in a distinct improvement of smoothness and accuracy of myocardial motion patterns over large displacements using time intervals of 201 ms (5 time points). Initial processing involved segmentation and tiling of the left ventricle. Deformable models were created for each time point of a single cardiac cycle. An initial triangular mesh gradually deforms to ®t every volume. Through this process regional excursions and speeds of each vertex representing a piece of myocardium can be calculated for each time-point interval throughout the entire cardiac cycle. Since the algorithm is indented to be modality independent, it should also be applicable to MR-Tagging data. Such data would allow an evaluation of the accuracy of the surface point-tracking algorithm. The algorithm is computationally ef®cient, producing each functional structure map throughout the complete cardiac cycle in less than two minutes. To animate the entire heartbeat, with encoded functional mappings and trajectory of the motion for individual segments, less than three minutes is required on a SGI workstation. Thus the algorithm processes data in a time frame short enough for immediate data processing, after patient data acquisition, for potential clinical application on a routine basis. The calculated results can be visualized through model animations and specially formatted static images. Regional maximum speed and excursion of myocardium through the cardiac cycle can be statically displayed through color mapping. In addition the value of regional maximum speed and excursion of myocardium during the cardiac cycle can be displayed through color mapping. Using animations, the local myocardial speed changes can be visualized through color change on the cardiac surface during the cardiac cycle. Moreover, the trajectory of
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493
491
Fig. 10. Display of the speed map and pathway trajectories of selected voxels during myocardial contraction (dark) and relaxation (white) on normal (left) and infarcted (right) infarction myocardiumÐwhite shades on the myocardial surface depict regions of little or no motion, and gray shades show relative speed values of moving myocardiumÐthe deeper gray corresponding to higher speed and more rapid motion.
motion for individual segments of myocardium can be displayed. 5. Summary The ability to encode quantitative functional information on a dynamic beating heart enhances the diagnostic value of 4D images in the cardiac cycle. Information regarding myocardial wall mechanics conveyed by such parametric displays adds another dimension to the assessment of some cardiac diseases. The ability to measure and display the motion of the heart walls combined with local myocardial excursion and speed maps could improve diagnosis of myocardial abnormalities, including regional ischemia, infarcts, akinesis and dyskinesis, diastolic contrast ®lling defects and/or electrophysiologic disturbances in regional conduction and contraction patterns. Acknowledgements The author expresses thanks to Wei-te Lin for his help with the deformable surface reconstruction program, to Pat Lund for assistance with segmentation and Mike Wahl, MD and Jon J. Camp for expert advice and technical support.
References [1] American Heart Association. Heart and StrokeÐStatistical Update, 1999. [2] Feiring AJ, Rumberger JA, Reiter SJ, Collins SM, Skorton DJ, Rees M, Marcus ML. Sectional and segmental variability of left ventricular function: experimental and clinical studies using ultra fast computed tomography. Journal of the American College of Cardiology. 1988;12(2):415±25. [3] Prince JL, McVeigh ER. Motion estimation from tagged MR image sequences. IEEE Transactions on Medical Imaging 1992;11:238±49. [4] Park J, Metaxas D, Axel L. Volumetric deformable models with parameter functions: a new approach to the 3D motion analysis of the LV from MRI-SPAMM. Fifth International Conference on Computer Vission, 1995. p. 700±5. [5] McEachen JC, Meyer FG, Constable RT, Nehorai A, Duncan JS. A recursive ®lter for phase velocity assisted shape-based tracking of cardiac non-rigid motion. http://noodle.med.yale.edu/mceachen/iccv/ iccv.ps.gz [6] Croissile P, Moore CC, Judd RM, Lima JAC, Arai M, McVeigh ER, Becker LC, Zerhouni EA. Differentiation of viable and nonviable myocardium by the use of three-dimensional tagged MRI in 2-dayold reperfused infarcts. Circulation 1999;99:284±91. [7] Papademetris X, Sinusas AJ, Dione DP, Duncan JS. 3D cardiac deformation from ultrasound images. Proceedings of MICCAI 1999;1679:420±9. [8] Mor-Avi V, Vigon P, Koch R, Weinert L, Garcia MJ, Spencer KT, Lang RM. Segmental analysis of color kinesis images: new method for quanti®cation of the magnitude and timing of endocardial motion
492
[9] [10]
[11]
[12]
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493 during left ventricular systole and diastole. Circulation. 1997; 95(8):2082±97. Shi P, Robinson G, Duncan JS. Myocardial motion and function assessment using 4D images. Proceedings of IEEE Visualization in Biomedical Computing 1994 Proc SPIE 1994;2359:148±59. Shi P, Amini A, Robinson G, Sinusas A, Constable CT, Duncan JS. Shape-based 4D left ventricular myocardial function analysis. Shapebased 4d left ventricular myocardial function analysis. Proceedings of the IEEE Workshop on Biomedical Image Analysis, Seattle, WA. IEEE 1994;327:604±12. Friboulet D, Magnin IE, Mathieu C, Pommert A, Hoehne KH. Assessment and visualization of the curvature of the left ventricle from 3D medical images. Computerized Medical Imaging and Graphics 1993;4/5:257±62. Gorce JM, Friboulet D, Clarysse P, Magnin IE. Three-dimensional velocity ®eld estimation of moving cardiac walls. Computers in Cardiology 1994;0276-6547/94:489±92.
[13] Clarysse P, Jaouen O, Magnin IE, Morvan JM. 3D representation and deformation analysis of the heart walls from X-ray and MRI images. Computers in Cardiology 1994;0276-6547/94:657±66. [14] St John Sutton MG, Ritman EL. Effects of progressive reduction in coronary blood ¯ow on regional and global left ventricular contraction and relaxation during normal and increased afterload: a roentgen videometric study. Cardiovascular Research 1982;16(9):535±45. [15] Robb RA, Hanson DP. The ANALYZE software system for visualization and analysis in surgery simulation. In Computer Integrated Surgery. Cambridge, MA: MIT Press, 1995. [16] Robb RA. Three-dimensional biomedical imagingÐprinciples and practice. New York, NY: VCH Publishers, 1995. [17] Lin WT, Robb RA. Realistic visualization for surgery simulation using dynamic volume texture mapping and model deformation. Proceedings of SPIE, Medical Imaging 1999;3658:308±31. [18] Hoffman EA, Ritman EL. Invariant total heart volume in the intact thorax. Am J Physiol (Heart Circ Physiol 18) 1985;249:H883±90.
C.D. Eusemann et al. / Computerized Medical Imaging and Graphics 25 (2001) 483±493 Christian D. Eusemann received his Diplom Ingenieur Fachhochschule from the University of Applied Sciences of Jena, Germany in 1999. He is currently a Graduate Student working on his PhD in Biomedical Engineering at Mayo Graduate School. He was the recipient of a Carl Duisberg Society Research Fellowship Abroad for University of Applied Sciences Students in Germany and a Whitaker Foundation Student Travel Award for SPIE 2001 Image-Guided Procedures meeting. His research interest is in Cardiac Functional Imaging, and he has published four papers in this ®eld.
Matthias E. Bellemann studied Physics, Mathematics, and Computer Science at the University of Heidelberg, Germany. During his studies, he received a renowned scholarship from the German National Scholarship Foundation (Studienstiftung des deutschen Volkes). Having obtained his Diploma degree in Physics in 1989, he started his studies in Medicine at the Medical School of the University of Heidelberg. In 1992 he received his PhD degree in Biophysics jointly from the Max-Planck-Institute of Experimental Medicine and the University of Heidelberg. During his academic and professional education, Dr Bellemann joined the Max-Planck Institute of Nuclear Physics in Heidelberg, the European Molecular Biology Laboratory in Hamburg, and the German Cancer Research Center in Heidelberg. In 1997 he was appointed as a full Professor of Medical Physics at the University of Applied Sciences, Jena, Germany. He is currently Associate Dean for Research Affairs at the Department of Biomedical Engineering at the University of Applied Sciences. His research interests are focused on the development and application of advanced imaging techniques for quantitative mapping of functional biomedical data. He is a member of the International Society for Magnetic Resonance in Medicine, the German Roentgen Society, the German Society for Nuclear Medicine, and the German Society for Medical Physics. He is reviewer of several prestigious national and international journals. He has been and is principal investigator on several national research grants. He has authored or co-authored over 40 peer-reviewed journal articles in the ®eld of biomedical imaging. Dr Bellemann's research interests are in functional biomedical imaging techniques (MRI, PET, CT, and US), image analysis (registration, parameter mapping), signal processing (biostatistics, pharmacokinetic modeling), and application of these procedures in experimental and clinical trials. He has developed several speci®c clinical applications of these techniques, including carbon-13 MR spectroscopy for tumor characterization and monitoring, pharmacokinetic modeling of PET data for follow-up studies during therapy, 3D image registration in MR mammography, and electromagnetic ®eld calculation for imageguided transcranial magnetic stimulation. His research interests also include the development and evaluation of new inter-modality imaging techniques as well as the combined application of diagnostic and therapeutic procedures.
493
Erik L. Ritman is Professor of Physiology and Medicine and is the Ralph B. and Ruth K. Abrams Professor at Mayo Medical School. He graduated in Medicine at the University of Melbourne, Australia, in 1964 and obtained his PhD in Physiology in 1973 at the University of Minnesota. His main interest is development of imaging devices for quantitative evaluation of cardiovascular structure-to-function relationships. In 1977 he became the principal investigator of the National Institutes of Healthprogram project grant that funded the fabrication and evaluation of the DSR, a high speed 3D CT scanner for study of humans and large animals. He currently heads a multidisciplinary NIH research grant which supports development and applications of X-ray micro-CT methods for study of small rodents' organs and biopsies from humans and large animals.
Richard A. Robb received the BA degree in Mathematics in 1965, the MS degree in Computer Science in 1968, and the PhD degree in Computer Science and Biophysics in 1971, all from the University of Utah. He is currently the Scheller Professor in Medical Research and Professor of Biophysics and Professor of Computer Science in the Mayo Medical School and Mayo Graduate School. He is Associate Dean for Academic Affairs in the Mayo Graduate School. He is Director of the Biomedical Engineering Program and Director of the Mayo Biomedical Imaging Resource at Mayo Foundation/Clinic. He has been involved in the development and application of computer systems for processing, analysis, and display of biomedical image data for over 28 years. He is a member of the American Physiological Society, the Biomedical Engineering Society, the Institute of Electrical and Electronics Engineers, the Society of Photo-Optical Instrumentation Engineers, the American Association for the Advancement of Science, the Association for Computing Machinery, the National Computer Graphics Association and the International Society for Computer Assisted Surgery. He is editor, associate editor and on the editorial board of several prestigious international journals. He has been and is principal investigator on several NIH research grants and has over 300 publications in the ®eld of biomedical image processing, including ®ve books and 30 book chapters. He has patented several inventions related to display, manipulation and analysis of computer-generated medical images. He has directed development of comprehensive software packages, which provide advanced capabilities for multidimensional biomedical image visualization and analysis. These software packages are used in over 300 institutions around the world and have been licensed to several commercial companies. Dr Robb's research interests are in biomedical image visualization; image processing (segmentation, registration and classi®cation), image modeling; virtual reality; advanced software systems; and distributed high performance computing systems and networks for biomedical image analysis. He has developed several speci®c clinical applications of these techniques, including 3D image-guided neurosurgery for brain cancer and epilepsy, prostate cancer diagnosis and treatment, quanti®cation and treatment of coronary artery disease, catheter-based myocardial ablation, radiation therapy planning, craniofacial reconstructive surgery and computerized histological analysis. His interests also include design and evaluation of new-generation paradigms for biomedical visualization systems of the future, particularly for medical treatment planning, minimally invasive clinical procedures, computer assisted surgery, and medical education.