Accurate Three-Dimensional Reconstruction of Intravascular Ultrasound Data
Spatially Correct Three-Dimensional Reconstructions
Background The geometrical accuracy of conventional three-dimensional (3D) reconstruction methods for intravascular ultrasound (IVUS) data (coronary and peripheral) is hampered by the inability to register spatial image orientation and by respiratory and cardiac motion. The objective of this work was the development of improved IVUS reconstruction techniques.
Methods and Results We developed a 3D position registration method that identifies the spatial coordinates of an in situ IVUS catheter by use of simultaneous ECG-gated biplane digital cinefluoroscopy. To minimize distortion, coordinates underwent pincushion correction and were referenced to a standardized calibration cube. Gated IVUS data were acquired digitally, and the spatial locations of the imaging planes were matched to their corresponding coordinates. Image points were then transformed relative to their respective 3D coordinates, rendered in binary voxel format, resliced, and displayed on an image-processing workstation for off-line analysis. The method was tested by use of phantoms (straight tube, 360° circle, 240° spiral) and an in vitro coronary artery model. In vivo feasibility was assessed in patients who underwent routine interventional coronary procedures accompanied by IVUS evaluation. Actual versus calculated point locations were within 1.0±0.3 mm of each other (n=39). Calculated phantom volumes were within 4% of actual volumes. Phantom 3D reconstruction appropriately demonstrated complex morphology. Initial patient evaluation demonstrated method feasibility as well as errors if respiratory and ECG gating were not used.
Conclusions These preliminary data support the use of this new method of 3D reconstruction of vascular structures with use of combined vascular ultrasound data and simultaneous ECG-gated biplane cinefluoroscopy.
High-frequency IVUS is a technique that allows visualization of in vivo vascular structures.1 2 3 This technique uses indwelling catheters with diameters as small as 1 mm with distally positioned 20- to 40-MHz piezoelectric transducers. Transverse cross-sectional images of peripheral and coronary arteries are obtained by advancing the 2D imaging plane at the catheter tip into an arterial segment to be visualized. This technique produces high-resolution images that have excellent structural definition of lumen and wall.
In vivo and in vitro validation of intravascular transducers has been performed primarily with straight vascular segments.4 5 6 Evaluation of the morphology and composition of atherosclerotic lesions within straight segments correlates well with pathological specimens.7 The utility of IVUS has been demonstrated during vascular interventional procedures to characterize atheroma size and composition, arterial geometry, results of interventions, and severity of dissections.8 9
There are, however, technical limitations to the procedure.10 IVUS images are obtained in a single plane perpendicular to the longitudinal axis of the catheter. Because only discrete 2D images are displayed at any given time, direct longitudinal vessel information and relationships are not displayed. Consequently, the operator is required to mentally integrate a series of transverse images to determine spatial relationships. The length of lesions and distance between landmarks can only be inferred by movement of the imaging plane.
To overcome the obstacles of 2D IVUS imaging, 3D reconstructions have been generated as a way of providing longitudinal information.11 12 The 3D display format provides a convenient way of integrating the 2D ultrasound data and appears to provide useful information concerning pathology, including dissections after intervention.13 Conventional 3DR algorithms, however, do not provide accurate spatial information. These approaches assume that the IVUS images were acquired from a straight vascular segment as parallel tomographic slices having a known separation distance. 3DR involves volumetric interpolation between adjacent imaging planes to fill in longitudinal image information. Since the sequence of input image files is reconstructed around a fixed, straight axis, the reconstructed vessels are displayed without curves. The slice separation distance is determined by the time required to traverse the segment during catheter withdrawal. With variation in pullback rate, vessel segments may be erroneously displayed as elongated or foreshortened. In addition, Waligora et al14 demonstrated that linear 3DR methods introduce substantial catheter-dependent geometric error in vessels with small radii of curvature.
Our purpose was to develop a method of 3D vascular reconstruction that used simultaneous acquisition of IVUS images and 3D ultrasound transducer coordinates. This method of 3D point registration was based on work originally performed by McKay et al to track left ventricular wall motion15 and augments recent work by Klein et al16 and Slager et al.17 Our method uses a calibrated spatial transformation algorithm applied to ECG-gated, biplane, digital, cinefluoroscopic images to determine the 3D coordinates of an IVUS catheter within the field of view. Gating allows these coordinates to be matched to the appropriate 2D ultrasound image for subsequent spatial orientation and rendering. We then tested the feasibility, accuracy, and clinical applicability of this method to assess accurate 3DR with IVUS data.
The IVUS catheter (Cardiovascular Imaging Systems, Inc) incorporates a 30-MHz single-element transducer with a rotating mirror enclosed in an acoustic housing at the distal end of the catheter. The mirror is located 1.8 cm proximal to the distal end of the catheter. The catheter is 135 cm in length and is advanced into position over a 0.014-in guide wire. The maximum outer dimension of the catheter is 5F (1.67 mm). Effective imaging is possible from the surface of the catheter to a depth of 3.2 mm. Lateral resolution of this system is 0.18 mm and axial resolution is 0.07 mm. Nominal two-point structural resolution is 0.1 mm at a focal distance of 2.2 mm.
A 1000 cm3 (10×10×10 cm) acrylic resin calibration cube with 27 embedded steel markers was constructed. The distance from the center of each marker to a reference corner (defined as the origin) was measured along each axis. Measurements were made in triplicate with machinist’s calipers and averaged to determine the known (x,y,z) coordinate locations of each marker (Fig 1A⇓).
A circular phantom was constructed by wrapping a 3.7-mm (OD), 2-mm (ID) guiding catheter segment around a 77-mm-diameter plastic film reel. A straight phantom was constructed by mounting another 3.7-mm (OD) guiding catheter segment to a rigid wooden dowel. A spiral phantom was constructed by spiraling a 4.0-mm (OD) polyethylene tube around a 24×90-mm cylinder. Before determining in situ accuracy of the arterial reconstruction, an ex vivo sheep coronary artery bifurcation was pressure fixed in formalin and used as an additional vascular phantom. Wire markers were placed on the phantoms to provide external, radiopaque landmarks. The distance between landmarks was measured in triplicate with calipers and averaged.
In Vitro Imaging
The calibration cube was imaged with biplane digital fluoroscopy (Coronix Bi-plane System with digital cardiac imaging, Philips Medical Systems, Inc) in triggered mode (Fig 1B⇑ and 1C⇑). The image intensifiers remained stationary and were positioned 90° from each other for all image acquisitions. If other orientations were used, a separate calibration of the image intensifiers had to be performed. The phantoms and vessel segment were imaged during pullback of the IVUS catheter. IVUS data were acquired simultaneously with the triggered digital biplane fluoroscopy. The fluoroscopic and ultrasound data were matched by gating the x-ray to an ECG simulator and simultaneously embedding the pulse signal on the ultrasound image, which was stored on videotape. Twenty to 31 paired biplane, digital, fluoroscopic images were obtained during manual pullback of the IVUS catheter through each phantom. The gated, paired x-ray images were transferred from the digital memory to hard-disk storage by use of a screen-digitizing program and network (CathView, ImageComm Systems).
Spatial Orientation From Biplane Projections
Spatial orientation is defined as the orientation of a 2D plane around a specific point in space. Our method of determining the spatial location (the position of a point in space) of an IVUS catheter tip from biplane cinefluoroscopic projections was adapted from the procedure described by McKay et al.15 This procedure involves two fundamental stages: (1) transformation matrix calculation from points having known spatial location and (2) 3DR of arbitrary points having unknown spatial location. As used in our application, the result of this procedure was accurate spatial location of the IVUS imaging plane (the 2D plane in which the IVUS data lay) in 3D space. This information also defined the centerline along which the discrete IVUS images were oriented. The spline-fit centerline is defined as the trajectory of the catheter that is a curve fit through a series of discrete R-wave–gated catheter mirror locations. Determination of the spatial orientation of the ultrasonic image required rotation of the images around the vessel centerline.
Briefly, the relationship between a 3D point (x,y,z) and the appearance of this point in an arbitrary 2D projection (u,v) is given by the equation:
where k is a scaling factor, T is a 4×3 transformation matrix that describes an arbitrary perspective view (eg, biplane view), and (x,y,z1) and (u,v1) are cartesian coordinate representations of the 3D and 2D points, respectively. In the current application, a unique view transformation matrix (T) was computed for each of the orthogonal cinefluoroscopic views. To determine the transformation matrix associated with the biplane views, the calibration cube (which possesses steel markers of known spatial location) is simultaneously imaged from both perspectives with the setup used to acquire the phantom or patient data. 2D-Projected positions (u1,v1) of the points of calibration object are related to their respective known 3D locations (x1,y1,z1). Each point (1) must satisfy the equation given above. The matrix equation results in a series of three linear equations in which the elements of T are the unknowns. Since the u,v terms can be measured from the projected biplane views and the x,y,z terms are known from the calibration cube, the 12 elements of the view transformation matrix (T) can be determined. For additional details, refer to McKay et al.15
The resulting view transformation matrix (T) allows a mathematical means of mapping arbitrary points with unknown 3D location in a given biplane (2D) projection to their actual position in 3D space. A minimum of 6 markers are required for accurate position identification; however, for the purpose of the present study, 15 markers were identified.
After the transformation matrix for both biplane views has been independently determined, the 3D spatial coordinates (x,y,z) of an arbitrary point (point associated with the imaging plane of an IVUS catheter) can be reconstructed from the measured location of the projected 2D points (u1,v1) and (u2,v2) in the two biplane views. This process is known as 3D point reconstruction and involves the least-squares solution of simultaneous matrix equations in which the only unknown quantities are x,y,z (Fig 2⇓).
Each ECG-gated ultrasound image was manually captured from videotape by use of the digitizing program on a commercial 3D workstation (StatView, ImageComm Systems). The raw IVUS image files of the phantom/vessel structure were processed via gray-scale thresholding and edited to remove remaining catheter artifact and noise by use of a commercial graphics utility program (OmniView version 2.1, Pura Labs). Because the value chosen for the threshold level plays a role in the 3DR algorithm, gray-scale thresholding was performed by a single operator (J.L.E.) on all studies. To establish the clinical reproducibility of our thresholding technique, intraobserver and interobserver variability was performed on 16 IVUS images by three blinded observers (M.J.V., W.B.B., and S.G.W.). The resulting image files were stored on a network file server (80386-based PC with Novell Advanced Netware 286 software). Fig 3⇓ demonstrates an image before and after processing.
The two transformation matrices (view 1 and view 2) were determined from the cinefluoroscopic views of the calibration cube (for matrix transformation calculation). Subsequently, the location of the IVUS transducer mirror in each gated/paired cinefluoroscopic frame of the pullback was identified by manually positioning a cursor over the IVUS mirror and determining the corresponding pixel address (Pixie version 1.5, ImageComm Systems) (Fig 4A⇓). We defined the location of the IVUS imaging plane as being centered on the midpoint of the rotating mirror. For images in which the catheter was rotated around the central axis, rotation correction was added. As catheter rotation was inhibited in our study, it was assumed to be negligible. A second-order polynomial correction was then performed to correct for pincushion distortion by use of a previously described method.18 Pincushion correction was performed once for the calibration cube and matrix and once for the image data in which the transformation matrix was used.
The processed 2D image data set was then converted from ImageComm format to PC-Matlab format (PC-Matlab version 3.5, The Math Works, Inc). Each IVUS image was then scaled, translated to its relative 3D position by use of the previously determined transformation matrices procedure, and rotated to a plane perpendicular to the centerline as determined by the best spline-fit curve of the sequence of IVUS transducer mirror location data. This spline-fit curve effectively represented the trajectory of the catheter during withdrawal. The imaging plane of the IVUS catheter was defined as the plane perpendicular to the longitudinal axis of the catheter shaft passing through the center (midpoint) of the acoustic mirror. The midpoint of the acoustic mirror was visually identified in each of the biplane fluoroscopic views by a human operator. Fig 4B⇑ illustrates graphically the 3D orientation of the data set before and after reorientation. Linear interpolation was performed between adjacent image sets, resulting in a volumetric (voxel) data set. The voxel data set was resliced along the z axis, creating a new series of parallel, 2D frames that were then reconverted to ImageComm format.
3D images were created by processing the data with algorithms developed specifically for voxel-based image display (Sonoview, Pura Labs). The resulting images were displayed on a workstation for display and analysis. Fig 4C⇑ illustrates the reconstructed images rotated in several planes.
Initial validation. The feasibility and accuracy of the 3DR method was tested with use of phantoms and an isolated sheep coronary arterial segment. Each preparation was imaged in a water bath and subsequently reconstructed. The phantom shape was compared qualitatively with the reconstructed shape. The distances between markers on the phantoms were determined by processing the x,y,z coordinates of the IVUS transducer mirror at the points (P1 and P2) where it crossed the planes of two marker wires by use of the following equation:
where P1=(x1,y1,z1) and P2=(x2,y2,z2).
Volume measurements were calculated from the phantoms by summing the product of the known 2D cross-sectional lumen areas and the distance between center points determined by use of the above equation (Simpson’s method). For the coronary specimens, histological volume measurements were computed as the product of marker distances and lumen areas determined from histological slides. The corresponding volume from the computer-reconstructed segment was determined as the product of the number of voxels that composed the lumen and the unit volume of each voxel.
Validation—ex vivo specimens. Seven calf coronary arteries were studied in situ, ex vivo. The coronary arteries were cannulated with 9F guiding catheters that were then fixed in position. The arteries were perfusion fixed with formalin at distending pressure (80 mm Hg), then the entire heart was fixed in formalin for 24 hours. After fixation, the coronary arteries were imaged in a water bath simultaneously with IVUS and with biplane digital cinefluoroscopy with use of an ECG simulator for gating. Data from one right, four LAD, and two circumflex coronary arteries were collected with the angiographic position registration system.
After imaging, the arterial segments were filled with a barium, formalin, and gelatin mixture at physiological distending pressure to preserve 3D morphology of the in situ vasculature. The arterial segments were then dissected from the hearts. Histological sections were made at 1.0-mm intervals by use of previously implanted markers on the arterial segments for reference to the 3D IVUS data.
Volume estimates of the coronary specimens were made by summing the product of histologically determined lumen area and the linear intersection distance (Simpson’s method).
Distance and volume measurement variability of the histological specimens previously has been shown in our laboratory to be <5% with this technique (unpublished data, 1984).
Initial clinical studies. This clinical protocol was approved by the Institutional Research Committee of Northwestern University. The feasibility of this method of 3DR was tested in seven patients (12 vessels). After obtaining informed consent, intracoronary ultrasound imaging was obtained in the cardiac catheterization laboratory after or during interventional procedures. In addition to routine intracoronary ultrasound imaging, a pullback through the vessel was recorded during a period of suspended respiration. Biplane digital fluoroscopy triggered by the patient’s ECG signal was performed simultaneously with the pullback. The ECG pulse signal was superimposed onto the ultrasound videotape by use of the technique described earlier. The position of the image intensifiers and the table were recorded and fixed during IVUS catheter pullback. As catheter mirror position was used to determine image orientation, the catheter was maintained in a stable rotational position throughout the pullback. After the study was completed, the image intensifier positions were reproduced and the calibration cube was imaged.
Because the data were continuously matched pairs, linear regression and correlation were used to describe the data sets. Results are expressed as mean±SEM. Paired t tests were used to determine whether there was a consistent error between the actual versus calculated measurements. A value of P<.05 was defined as significant for all comparisons.
The purpose of the initial phase of the project was to develop and test our algorithms.
Fig 4⇑ demonstrates the various stages of our testing with a circular phantom. Fig 4A⇑ represents two views at two time sequences of the original phantom with the ultrasound transducer present within the tube lumen. Fig 4B⇑ illustrates our spatially oriented ultrasound data before and after reorientation of the image planes. Fig 4C⇑ illustrates the final 3DR after interpolation and smoothing of data points. The marked similarity of shape and proportion is easily noted by comparing Fig 4A⇑ and 4C⇑. The opening in the circular reconstruction represents the start and end of the transducer pullback.
Fig 5⇓ depicts the reconstruction of the spiral phantom. Fig 5A⇓ illustrates the digital cinefluoroscopic image of the contrast-filled phantom and Fig 5B⇓, the corresponding 3D IVUS reconstruction. The shape and proportion of the 3DR are similar to those of the phantom.
We were able to demonstrate with a variety of phantoms that these algorithms could be used to make recognizable reconstructions that were qualitatively geometrically similar.
Phantoms. To aid in quantifying our data, radiopaque markers were placed on each of our phantoms as noted on Figs 4A⇑ and 5A⇑. Table 1⇓ lists and Fig 6⇓ illustrates the measured intermarker distances on each of the phantoms (straight, circular, spiral, and coronary bifurcation) compared with the calculated distances derived from the transducer position registration data. Twenty-one chords and diameters were measured on the circular phantom. Ten distance measurements were obtained from the straight phantom, four from the spiral phantom, and four from the sheep coronary artery preparation. Actual versus calculated measurements were very close. All errors were within 2 mm for segments shorter than 4 cm and within 5 mm for the longer segments (P=NS). Volumes were determined for each of the tubes used in phantom reconstruction. Table 2⇓ demonstrates the close comparison of the measured volume of each tube to our calculated volume. The calculated errors were within 3%.
In vitro specimens. Fig 7⇓ illustrates comparisons of luminal distances, areas, volumes, and total volumes for the in situ, ex vivo vascular segments. Total volume was defined as luminal plus wall volume. The arterial wall was defined by IVUS from the leading edge of the inner bright specular reflector to the trailing edge of the outer bright specular reflector that subtended the arterial segment and by histology from the intima to the dense adventitia. There was good correlation between IVUS 3DR and histological (HISTO) measurements, as follows:
Luminal distances (mm): 3DR=0.9 HISTO−0.6; r=.84; n=22.
Luminal areas (mm2): 3DR=1.0 HISTO+1.0; r=.88; n=27.
Luminal volumes (mm3): 3DR=0.9 HISTO+4.3; r=.81; n=21.
Total volumes (mm3): 3DR=0.9 HISTO+35.0; r=.83; n=21.
Intraobserver and interobserver variability of the thresholding border technique was found to be 8.1% and 9.8%, respectively. These results were based on three blinded reviewers who each set thresholds twice on 16 representative IVUS images.
A total of 12 vessels (native coronary arteries and saphenous vein grafts) were imaged and reconstructed in seven patients. All patients were undergoing an interventional coronary procedure at the time of IVUS collection. The time added for data collection was generally <4 minutes (maximum for 2 vessels was 10 minutes), with actual imaging lasting 20 to 30 seconds. Fig 8⇓ depicts an angiogram of a LAD coronary artery and the corresponding reconstruction. The reconstruction qualitatively depicts vascular geometry and the proximal curve.
Gating for ECG and respirations was important to avoid introducing motion artifacts into the reconstruction. Fig 9⇓ illustrates an example of a saphenous vein graft angiogram (Fig 9A⇓) and 3DR (outer wall, Fig 9B⇓) distorted by breathing. The angiogram demonstrates no tortuosity.
Of the 12 vessels that were initially reconstructed, 7 visually demonstrated luminal geometry similar to the angiographic segment (1 proximal LAD; 2 mid-LAD coronary arteries; 1 first obtuse marginal; and 3 saphenous vein grafts to a LAD, right coronary, and first obtuse marginal, respectively). Respiratory motion resulted in distorted 3DR of 3 segments (1 saphenous vein graft, 1 obtuse marginal, and 1 right coronary). IVUS image dropout hampered good 3DR of 2 saphenous vein grafts (to a diagonal and right coronary artery). As our present thresholding algorithm does not add missing information, image dropout was demonstrated in the reconstructions if present in the original data.
Overall, if respiration motion was suspended and image borders could be detected on the IVUS data, the 3DR visually described the relationship of each segment as displayed by the angiogram.
Additional unexpected information was found. Fig 10⇓ illustrates a 3DR (Fig 10B⇓) of a saphenous vein graft to an obtuse marginal branch with the angiogram on the left (Fig 10A⇓). Because the saphenous vein graft is untethered, there is straightening of the vessel due to the presence of the catheter in the graft during pullback through the curve (arrow in Fig 10B⇓). This type of logical information would be missed with traditional 3DR techniques.
IVUS provides in vivo information regarding vascular morphology and pathology. The present utility of IVUS is limited by the display format, which demonstrates only single transverse or oblique vascular sections at a single time and provides only indirect longitudinal information.
Our data has demonstrated the accuracy and initial clinical applicability of a method that allows spatially correct 3DR and display of IVUS data in the coronary bed. The discussion will focus on previous methods of 3DR, advantages of our technique, potential sources of error, and applications.
Previous Methods of 3DR
Studies have demonstrated the feasibility and benefits of 3DRs of vascular segments.11 12 13 Despite anatomic limitations, clinically useful information can be obtained with near real-time reconstruction of IVUS images acquired in the catheterization laboratory. The resulting reconstructions appear straight but can be rotated around each axis and electronically sectioned to reveal internal structure. These images can visually provide an estimate of lesion length, dissections, and vessel segment morphology. Although this technique may be useful in many peripheral vascular beds, which are relatively straight, or in very small coronary artery segments, inherent tortuosity of the vasculature degrades the geometric accuracy of conventional 3D IVUS. Atheroma provides further tortuosity and may add to the inaccuracy of nonspatially correct 3DR.10 Klein et al16 described a method that used biplane angiography to define the vessel midline in 3D space. Serially, collected IVUS images from the target vessel were aligned perpendicular to the vessel midline, subjected to longitudinal interpolation, and displayed as a 3D structure. Recent work by Slager et al17 extended the concept of combining biplane angiography and IVUS for vascular 3DR. Their technique involved estimation of the trajectory of the IVUS catheter. Corresponding image data were then arranged perpendicularly around the trajectory axis, and 3DR was performed. Both methods16 17 produced accurate renderings of vascular geometry and are promising analogues to the method we report. Attributes of our method are the inclusion of cardiac and respiratory gating to minimize 3DR motion artifacts, the extension into the coronary bed, and extensive validation.
We reviewed a total of 17 different techniques that described 3DR of vascular and ventricular structure with IVUS, B-mode ultrasound, and radiographic modalities. The method that we chose was originated by McKay et al.15 Our adaptation of this technique is unique and applicable when ease of implementation, compatibility of the technique with our facilities, and the efficacy of a least-squares error fit are considered. It does not require other resources except a calibration cube and gated biplane cineangiography.
McKay’s algorithm15 was never intended to provide 3DR of the coronary tree, although he did estimate the geometry of these structures in schematic fashion. Rather, he was interested in cardiac wall motion ascertained through 3D motion branch points of the coronary circulation. However, this algorithm, as we have demonstrated, can be well adapted for the vascular bed.
Our reconstruction technique has several potential advantages. By using 3D coordinates, we can track the position and vector of the transducer in 3D space. This allows placement of the ultrasound image data into a 3D matrix in a position that is representative of its true position and spatial orientation.
Cardiac and respiratory motion influence our results and are the primary reason for ECG gating and respiratory suspension. The time required to implement the spatially correct algorithm is ≈4 minutes per vessel added to the cardiac catheterization procedure. Additionally, 15 minutes is required for setup before the procedure and 60 minutes for the 3DR. More sophisticated processing equipment could further reduce the 3DR time. In fact, with current trends in computing power, on-line data collection, orientation, and display should be achievable.
The mean ultrasonic image slice distance depends, to some extent, on the pullback time. With our very slow manual pullback and use of centerline geometry, our distance is presently 1 mm. If a much slower pullback occurred, the number of image slices per millimeter could be increased. Newer, motorized pullback devices allow IVUS data collection at constant pullback rates as low as 0.5 mm/s. Automated pullback devices allow a standard, uniform pullback of the catheter at the proximal:distal end. In tortuous vascular beds and especially in the coronary beds, a linear pullback 1:1 ratio cannot be obtained even with use of these automated pullback devices. In addition, accordioning (vessel peel off) as the catheter is pulled back causes further difficulty in systems that require uniform pullback for 3DR. Our algorithm does not require the assumption that 1:1 matching has occurred. Recent development of sheath-type IVUS catheters, in which the IVUS imaging element can be withdrawn through the body of the catheter shaft while the catheter itself remains stationary, may improve the reliability of timed pullback techniques, reduce accordioning distortion, and improve 3DR IVUS accuracy.
A major problem with present linear reconstruction techniques, especially in the coronary bed, is their lack of true image orientation. Research conducted by our laboratory suggests that there is an inverse relation between vessel curvature and the volumetric accuracy of 3DR IVUS.14 Despite the use of slow, motorized catheter withdrawal and sheath-type IVUS catheters, important errors may be incorporated into 3DR IVUS techniques that do not account for the true image orientation. This is displayed graphically in our example (Fig 10⇑) of the saphenous vein graft 3DR that, by angiography, has one small curve but that demonstrates a much larger iatrogenically induced curve when the catheter is pulled through the vascular segment.
We have demonstrated with a variety of phantoms that our reconstructions are representative in shape and accurate in position. Since we use this method to position the long axis of the IVUS catheter, which will move several centimeters, an error introduced by this technique should be within an acceptable range.
Potential Sources of Error and Study Limitations
Several factors were noted that influenced our results. Our in vitro vascular imaging results were limited by movement of the arterial segment in the water bath during pullback, with accordioning (vessel peel off) of the segment over the catheter. This movement artifactually caused a systematic error in the distance calculations determined from the foreshortened segment when compared with actual measurements made with the artery fully extended. Similarly, respiratory suspension limited this study to short and medium-length arterial segments. Newer techniques of respiratory gating should allow accurate 3DR of relatively long segments. Our cube, although custom made, was not precision instrumented. The accuracy of our transformation matrix should improve further with a precision-instrumented cube. We determined all of our x,y,z coordinate locations by using the pixel address of the point on digitized images. The pixel density was 512×480 for a 9-in digital fluoroscopic image. The higher line density (1024×1024) of many angiographic monitors should increase accuracy. Likewise, the process of manually identifying the catheter mirror in the fluoroscopic images was a source of variability. Scaling was required to display the entire reconstruction on our video screen, and this necessitated reduction of ultrasound image data points for reconstruction. Although this data reduction clearly decreases the quality of the ultrasound image, it is likely that continued improvements in computing speed and memory will necessitate less scaling in future adaptations. Full gray-scale resolution, which provides additional visual cues regarding vessel composition, was difficult with our present image processing system. Full gray-scale 3DR is available on many newer image processing systems and will allow better identification of plaque load and atheroma components.
The results in Fig 7⇑ demonstrate a better correlation of our technique to histology when lumen distance and area measurements are compared than when volume measurements are compared. The larger volume error can be attributed to the difference in methods of volume computation. Histological volume was computed as a direct product of lumen distance (ie, a straight line) and area, whereas the IVUS 3DR volume was calculated as the interpolation between each set of lumen areas by use of a spline-fit curve technique. This discrepancy in volume measurements is greater in tortuous and curved segments, in part because of curvature that cannot be accounted for with most methods of histological volume calculation, including our own.
An additional source of 3DR error was the presence of structural mirror artifacts (“strut” artifacts) in the 2D IVUS images. These artifacts are inherent in the rotating mirror catheter design used in the current protocol and should be reduced with newer-generation imaging devices. Although this artifact did not affect the accuracy of catheter coordinate locations, it could result in lumen and wall area and volume errors, because it tends to obscure and distort a sector of visualized arterial wall. This can be demonstrated by our variability measurements of the threshold technique for IVUS edge detection. Although intraobserver and interobserver variability both were within 10%, which is acceptable for ultrasound image data, the variability in measurements may be due to edge clarity of the IVUS image data. Further improvements in IVUS presentation of image data may decrease this variability. Our method used respiratory suspension, which limits the present method to short or moderate-length coronary and peripheral vascular segments. Newer methods of respiratory gating should allow adaptation of this technique to large arterial segments.
Despite these limitations, our results are very promising. Our study was primarily designed to test the feasibility of this technique for obtaining spatially correct 3DRs, to validate the technique, and to demonstrate feasibility in the clinical setting. A larger clinical study will be required to determine utility.
True 3DR of vascular studies will have multiple basic and clinical applications. The potential exists to quantify the volume load of atheroma that is present within a vessel segment. Atheroma removal or displacement of this material can be evaluated after an interventional procedure in either a coronary artery or a peripheral vessel. In addition, the evaluation of complications of interventions, such as the length and depth of a dissection, should be quantifiable.
There are numerous basic applications for this technique. Accurate geometric reconstructions of arterial wall and lumen will allow true evaluation of changes in segmental vascular reactivity with atheroma and hypertension and after intervention.
We have demonstrated that spatially correct 3DRs of IVUS data with simultaneous biplane-triggered fluoroscopy is possible. In phantoms, the reconstructions have been shown to appear correct, with accurate distances and volume calculations. We also have demonstrated the feasibility of applying this technique in patients at catheterization. With this method, the potential exists for accurate display and analysis of vascular wall and luminal anatomy.
Selected Abbreviations and Acronyms
|LAD||=||left anterior descending|
This study was supported in part by the Chicago Heart Association, the Feinberg Cardiovascular Research Institute, and the National Institutes of Health (grant No. HL-46550). The authors gratefully acknowledge manuscript preparation by Jean Waller and Cynthia Shane.
Reprint requests to James L. Evans, MD, Tucson Medical Center, 2355 N Ferguson Ave, Ste 111, Tucson, AZ 85712, or David D. McPherson, MD, Wesley - Ste 524, Northwestern Memorial Hospital, 250 E Superior, Chicago, IL 60611.
- Received April 24, 1995.
- Revision received September 12, 1995.
- Accepted September 17, 1995.
- Copyright © 1996 by American Heart Association
Nissen SE, Grines CL, Gurley JC, Sublett K, Haynie D, Diaz C, Booth DC, DeMaria AN. Application of a new phased array ultrasound imaging catheter in the assessment of vascular dimensions: a comparison to cineangiography. Circulation. 1990;81:660-666.
Tobis JM, Mallery J, Lehman K, Zalesky P, Griffith J, Gessert J, Moriuchi M, McRae M, Mahon D, Dwyer M-L, Greep N, Henry WL. Intravascular ultrasound imaging of human coronary arteries in vivo: analysis of tissue characterizations with comparison to in vitro histological specimens. Circulation. 1991;83:913-926.
Yock PG, Johnsen EL, David DT. Intravascular ultrasound: development and clinical potential. Am J Card Imaging. 1988;2:185-193.
McKay CR, Hahn L, Nuno D, Fleagle S, Kerber PE, Marcus M. In vivo validation of intraluminal ultrasound catheter measurements using sonomicrometers and quantitative angiography. J Am Coll Cardiol. 1990;15:16A. Abstract.
Isner JM, Rosenfield K, Losordo DW, Kelly S, Palefsky P, Langevin RE, Razvi S, Pastore JO, Kosowsky BD. Percutaneous intravascular ultrasound examination as an adjunct to catheter-based interventions: preliminary experience in patients with peripheral vascular disease. Radiology. 1990;175:61-70.
Roelandt JR, di Mario C, Pandian NG, Wenguang LI, Keane D, Slager CJ, deFeyter PJ, Serruys PW. Three-dimensional reconstruction of intracoronary ultrasound images: rationale, approaches, problems, and directions. Circulation. 1994;90:1044-1055.
Rosenfield K, Losordo DW, Ramaswamy K, Pastore JO, Langevin RE, Razvi S, Kosowsky BD, Isner JM. Three-dimensional reconstruction of human coronary and peripheral arteries from images recorded during two-dimensional intravascular ultrasound examination. Circulation. 1991:84:1938-1956.
Waligora MJ, Vonesh MJ, Wiet SP, McPherson DD. Effect of vascular curvature on three-dimensional reconstruction of intravascular ultrasound images. Circulation. 1994;90:I-277. Abstract.
Klein H-M, Gunther RW, Verlande M, Schneider W, Vorwerk D, Kelch J, Hamm M. 3D-Surface reconstruction of intravascular ultrasound images using personal computer hardware and a motorized catheter control. Cardiovasc Intervent Radiol. 1992,15;2:97-101.
Slager CJ, Laban M, von Birgelen C, Krams R, Oomen JAF, den Boer A, Wenguang L, deFeyter PJ, Serruys PW, Roelandt JRTC. ‘ANGUS’: a new approach to three-dimensional reconstruction of geometry and orientation of coronary lumen and plaque by combined use of coronary angiography and IVUS. J Am Coll Cardiol. 1995;25:144A. Abstract.
Tehrani S, LeFree MT, Sitomer J, Bourdillon PDV. High-speed digital radiography pincushion correction using an array processor. IEEE Trans Biomed Eng. 1987:615-618.