A New Method for Cardiac Computed Tomography Regional Function AssessmentClinical Perspective
Stretch Quantifier for Endocardial Engraved Zones (SQUEEZ)
Background—Quantitative assessment of regional myocardial function has important diagnostic implications in cardiac disease. Recent advances in CT imaging technology have allowed fine anatomic structures, such as endocardial trabeculae, to be resolved and potentially used as fiducial markers for tracking local wall deformations. We developed a method to detect and track such features on the endocardium to extract a metric that reflects local myocardial contraction.
Methods and Results—First-pass CT images and contrast-enhanced cardiovascular magnetic resonance images were acquired in 8 infarcted and 3 healthy pigs. We tracked the left ventricle wall motion by segmenting the blood from myocardium and calculating trajectories of the endocardial features seen on the blood cast. The relative motions of these surface features were used to represent the local contraction of the endocardial surface with a metric we call Stretch Quantifier of Endocardial Engraved Zones (SQUEEZ). The average SQUEEZ value and the rate of change in SQUEEZ were calculated for both infarcted and healthy myocardial regions. SQUEEZ showed a significant difference between infarct and remote regions (P<0.0001). No significant difference was observed between normal myocardium (noninfarcted hearts) and remote regions (P=0.8).
Conclusions—We present a new quantitative method for measuring regional cardiac function from high-resolution volumetric CT images, which can be acquired during angiography and myocardial perfusion scans. Quantified measures of regional cardiac mechanics in normal and abnormally contracting regions in infarcted hearts were shown to correspond well with noninfarcted and infarcted regions as detected by delayed enhancement cardiovascular magnetic resonance images.
- four-dimensional computed tomography
- cardiac imaging techniques
- magnetic resonance imaging
- ischemic heart disease
- myocardial contraction
- volumetric computed tomography
Coronary angiography is currently the most prevalent use of cardiac CT. In this study, we aimed to assess systolic regional cardiac function from high-resolution volumetric cardiac CT acquisitions, which can be acquired in conjunction with routine CT angiography. Assessment of regional myocardial function has value in the diagnosis and monitoring of myocardial ischemia and myocardial dyssynchrony.1,2 Most mechanical analyses in the clinical setting is based on echocardiographic methods derived from 2D motion data. Not all tomographic imaging modalities are capable of producing data with adequate temporal and spatial resolution for detailed regional function assessment. One difficulty with quantitative tomographic methods to estimate myocardial function is the inability to obtain adequate landmarks in the heart because of poor spatial resolution.
Clinical Perspective on p 250
Cardiovascular magnetic resonance (CMR) tissue tagging, which is currently considered the reference method, is validated and accurate, but it is slow, has poor resolution in the slice selection direction, and requires extended breath holding, and its image analysis is time consuming because of the manual segmentation required to detect the myocardial borders.3 In addition, CMR imaging is still considered a contraindication in the rapidly growing population of patients with implanted pacemakers or implantable cardioverter-defibrillators.
Recent dramatic advances in cardiac CT imaging techniques allow for volumetric functional imaging of the entire heart with a few gantry rotations.3–9 The high temporal resolution acquisitions of the entire cardiac volume with wide-range detector CT allows a contrast bolus to be imaged over a short window in the heart cycle with very high spatial resolution, making visible fine anatomic structures, such as trabeculae, on the endocardial surface. We took advantage of the resolution now available with wide-range detector CT to develop a method to detect and track the fine curvature-based geometric features on the endocardial surface, which are used to extract a metric that reflects the cardiac muscle contraction. It has been previously shown that differential geometry features of the myocardial surfaces can be used to estimate the motion field from 3D anatomic images10–12; however, the low spatial resolution of the images has been a limitation. Here, we evaluate the feasibility of tracking the left ventricular (LV) wall motion and assessing local cardiac function in high-resolution first-pass volumetric cardiac CT images using a fast, nonrigid, surface registration algorithm that matches geometric features of the surface over time.
All animal studies were approved by the Johns Hopkins University Institutional Animal Care and Use Committee and comply with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health Publication no. 80–23, revised 1985).
Pigs with chronic myocardial infarctions (MIs) were created as previously described.13 Briefly, MI was induced by engaging the left anterior descending coronary artery (LAD) with an 8F hockey stick catheter under fluoroscopic guidance. Then, a 0.014-in angioplasty guidewire was inserted into the LAD, and a 2.5×12-mm Maverick balloon (Boston Scientific) was inflated to 4 atm just distal to the second diagonal branch of the LAD. After 120 minutes, occlusion of the vessel was terminated by deflating the balloon, and restoration of flow in the LAD was confirmed by angiography. CT and MRI studies were performed ≈130 to 180 days after MI induction. A total of 11 animals were studied (7 chronic MI, 1 acute MI, and 3 healthy).
Each animal was scanned with electrocardiographic monitoring using a 0.5-mm×320-row detector scanner (Aquilion ONE; Toshiba Medical Systems Corporation). Animals received intravenous metoprolol (2–5 mg), amiodarone (50–150 mg), or both to achieve a heart rate of <100 beats/min. After scout acquisition, a 50- to 60-mL bolus of iodixanol (320 mg iodine/mL; Visipaque; Amersham Health) was injected intravenously, and a first-pass cardiac perfusion scan for the entire cardiac cycle was performed. During CT acquisition, respiration was suspended, and imaging was performed using a retrospectively gated CT protocol with the following parameters: gantry rotation time, 350 ms; temporal resolution, up to 58 ms using multisegment reconstruction9; detector collimation, 0.5 mm×320 rows (isotropic voxels, 0.5×0.5×0.5 mm3); tube voltage, 120 kV; and tube current, 400 mA. One infarcted data set was acquired using x-ray tube current modulation of 10% of the maximum, with the maximum current at only the 75% time point of the R-R interval. Images were reconstructed at every 10% of the R-R interval in systole using a standard kernel (FC03), QDS+ noise reduction filter, and a multisegment (3–5 beats) reconstruction algorithm. Electrocardiographic editing to account for arrhythmias was performed when necessary. In addition, a set of low-dose, prospectively gated scans (120 kV and 20 mA at 0% and 50% of R-R) along with a high-dose (120 kV and 400 mA) retrospectively gated scan were acquired for 1 animal to assess the feasibility of tube current reduction and prospective gating for cardiac function analysis.
Cardiovascular Magnetic Resonance
In vivo CMR images were acquired using a 3T MR scanner (Achieva; Philips) with a 32-element cardiac phased array. Myocardial viability was visualized using late gadolinium enhancement images acquired 20 to 25 minutes after intravenous injection of a double dose of gadolinium diethylenetriaminepentaacetic acid (0.2 mmol/kg body weight) (Magnevist; Berlex). A 3D, ECG-triggered, independent respiratory navigator-gated, breath-hold, phase-sensitive inversion recovery gradient echo imaging pulse sequence was used.14 Imaging field of view was 24×24×12 cm3, with an imaging matrix of 200×195×30, yielding an acquired voxel size of 1.20×1.23×4.0 mm3 reconstructed to 0.91×0.91×2.0 mm3. Other relevant imaging parameters were as follows: flip angle, 15°; repetition time, 5.3 ms; echo time, 2.6 ms; and receiver bandwidth, 289 Hz/pixel.
For each systolic cardiac phase, the blood in the LV was segmented from the myocardium by thresholding the voxel intensities roughly between 200 and 650 Hounsfield units. After manually pruning the coronaries; aorta; and, in some data sets, the right ventricle (using the Medical Image Processing, Analysis, and Visualization program available from the National Institutes of Health at http://mipav.cit.nih.gov), a triangulated mesh representing the endocardial surface was extracted from the boundary surface of the LV blood cast (Figure 1A–1C and 1E). All computations, unless specified otherwise, were done using Matlab (MathWorks Inc) software. To compare the results of the proposed algorithm to existing CT wall motion tracking software, the data sets were analyzed using Vitrea fX software (Vital Images).
We tracked the LV wall motion by calculating trajectories for the points on the endocardial mesh. Each endocardial surface was represented by a triangular mesh. To track the points on the meshes from end diastole (ED) to end systole (ES), the surfaces needed to have the same number of triangles, with a 1:1 correspondence between the vertices. This was accomplished by choosing a template mesh (in this case, the ED mesh) and warping it onto a target mesh (any systolic mesh, eg, the ES mesh) such that every triangle on the template mesh had a corresponding triangle on the target mesh (Figure 1D and 1E). We chose a nonrigid point registration algorithm termed coherent point drift (CPD) for surface warping. CPD is a probabilistic method used for nonrigid surface registration in which surface points are forced to move coherently as a group to preserve the topological structure of the point sets.15 The coherence constraint was imposed by regularizing the displacement field and using variational calculus to derive the optimal warping. A fast implementation of CPD, based on the fast Gaussian transform, was used to reduce the massive computational burden associated with high-resolution CT data.
To match the anatomy through surface warping, the homologous anatomic features and their correspondences needed to be identified. Therefore, features engraved on the endocardial surface by fine anatomic structures, such as trabeculae and papillary muscles, were encoded using a scale-independent local shape measure termed shape index (SI) (Figures 1D and 2) and incorporated in the warping algorithm to further improve its accuracy. The SI is a curvature-based measure, and for each point is defined by (1) where k1 and k2 are the principal (signed maximum and minimum) curvatures at that point. Figure 2 shows SI values for different surface shapes. For a saddle point, k1=−k2; thus, SI=0. For a spherical surface, k1=k2≠0, and the SI=−1 if curvatures are negative and +1 if they are positive, corresponding to a spherical cup and cap, respectively. For a valley, k1=0, and k2 can have any negative value (by definition k1≥k2); thus, as long as k2 is nonzero we have:
An important property of SI is that it is stretch invariant. As mentioned previously, surface features (eg, ridges and valleys) will have a certain SI value solely based on their shape and not on their curvatures (ie, steepness). Therefore, as long as the topology of the surface does not change under compression or stretch, the anatomic features, such as ridges and valleys on the endocardial surface, will retain their SI values. This property makes SI a useful tool for encoding endocardial features.
The output of the CPD algorithm is a displacement field that is used to calculate measures of local cardiac function. We propose a measure of local cardiac function called Stretch Quantifier of Endocardial Engraved Zones (SQUEEZ), which is defined as follows: (2) where A(v, 0) is the area of the small triangular patch (v) on the endocardial mesh at ED, and A(v, t) is the area of the same patch at time t. SQUEEZ is calculated for each triangular patch on the surface, resulting in a high-resolution local cardiac function map of the LV.
For the data pool obtained from the 11 animals, 2-tailed paired Student t test statistical analyses were performed on the SQUEEZ value and the slope of SQUEEZ versus time to assess the difference in the means of these parameters in healthy and infarcted regions. The accuracy of the registration algorithm was evaluated using the mean of the minimum pairwise Euclidean distance between the target and the warped data sets (ie, for each point on the template mesh, the Euclidean distance to every point on the warped mesh is calculated, and the minimum is chosen). The mean±SD of the minimum distances is reported.
To evaluate resting LV function, the blood pool of the LV was segmented in the ED and ES phases in the 3D volume and ED volume, ES volume, stroke volume, and ejection fraction were calculated for the LV (Figure 3). SQUEEZ values were measured in healthy and infarcted animals at different cardiac phases (Figure 4) and different locations of infarcted and remote myocardium as detected by contrast-enhanced MRI (Figure 5).
Accuracy of the Nonrigid Registration Algorithm (CPD)
The accuracy of the nonrigid registration algorithm was evaluated using the mean of the minimum Euclidean distance between the target and warped surfaces evaluated at all points. Over the 11 animals analyzed by our method, there was a subpixel average error of 0.6±0.4 pixels (0.3±0.2 mm). All the triangular patches on the meshes had sides ≥1 pixel.
Regional Cardiac Function
SQUEEZ was calculated for every point on the LV endocardial surface at each cardiac phase. All infarcted animals showed abnormal stretching in the LAD territory, which was consistent with the infarct model used in this work. One animal showed 2 distinct MI zones, and this was confirmed by examining the CMR image, which showed a secondary MI in the inferior wall.
Figure 4 shows SQUEEZ bull's-eye plots calculated for 5 consecutive phases from ED to ES. Areas in yellow (SQUEEZ >1) show abnormal stretch because of MI.
Contrast-enhanced CMR images were used as the gold standard to verify the location of the infracted regions detected in SQUEEZ maps (Figure 5A). Points were selected on regions of the endocardial surface near the MI zones as defined by the contrast-enhanced CMR images. Approximately the same number of points were selected in a remote region of the heart with no sign of MI (Figure 5B). The size of the selected regions roughly corresponded to that of 1 LV segment in the 17-segment American Heart Association model.18
The average SQUEEZ value was calculated for each zone and showed a significant difference (P<0.0001) between MI and non-MI regions in infarcted animals (Figure 5B). For healthy animals, a region on the lateral wall was chosen corresponding to the remote non-MI region selected in infarcted animals. The SQUEEZ values for the non-MI region in the infarcted hearts and the regions chosen in the healthy hearts were not significantly different.
In addition to SQUEEZ, the rate of change in SQUEEZ also showed a significant difference (P<0.0001) between MI and non-MI regions in the infarcted animals (Figure 5C), and no difference was found between the same lateral regions in healthy and non-MI regions. Non-MI regions showed an average SQUEEZ rate of ≈−0.6± 0.2, whereas the MI zones had a rate of ≈0±0.1, showing little or no stretch or contraction.
The SQUEEZ time plots for the tube current modulated data set showed higher SDs because of increased noise levels. However, the difference between MI and non-MI regions was still significant, and the trend of the plots were similar to those of the high-dose data sets (Figure 5B).
The SQUEEZ map was calculated for the low-dose prospectively gated data set and compared to the SQUEEZ of the high-dose retrospectively gated data set at 50% of the R-R interval. The difference between the SQUEEZ maps was computed (Figure 6). The results show low bias (0.01; 95% CI, −0.12 to 0.15) between the high-dose retrospective and the low-dose prospective scans. The differences could be attributed not only to the increased noise due to lower tube current, but also to heart rate variations among the acquisitions. More experiments are going be carried out to fully investigate the effects of CT noise on the accuracy of SQUEEZ. Use of the low-dose prospective scan decreased the radiation dose by ≈10-fold. The low bias and 95% CI of the low-dose scan make the use of low-dose, prospectively gated CT for cardiac function very promising.
Regional ejection fraction (rEF) was calculated at ES for each cardiac segment using Vitrea fX software. The automatic segmentation of endocardial borders required manual correction, which took ≈150±15 minutes, as opposed to 4±2 minutes of operator interaction required in the proposed method. SQUEEZ values were averaged into the American Heart Association 16 segments and compared to 1-rEF values obtained from Vitrea fX. There was good correlation (r=0.81, P<0.001) for the 6 midcavity segments (segments 7–12), but no correlation was found in basal and apical segments in any of the data sets.
We have developed a new method for measuring regional cardiac function with high resolution from volumetric CT acquisitions that has proven to be effective in quantifying regional cardiac mechanics and detecting infarcted regions in a large animal model. Volumetric CT data used in this work can be reconstructed from the routine dose-modulated coronary angiography CT scans. The method also eliminates the laborious human interaction required to segment the cardiac data for functional analysis that has plagued cardiac imaging for the past 2 decades.
Current methods for CT regional cardiac function analysis involve time-consuming manual segmentation or manual correction of segmentation of the myocardium. These methods generally apply smooth contours to delineate epicardial and endocardial boundaries, thus failing to capture the fine anatomic endocardial structures visible in wide-range detector CT that can be used as landmarks to guide the motion tracking algorithm. Furthermore, these algorithms normally track the 3D motion by analyzing the displacements in stacks of 2D slices (usually oriented in the short axis). The longitudinal displacement of tissue into and out of a short-axis slice can appear to be a change in the myocardial wall thickness in that short-axis slice, but it is in fact just bulk displacement of tissues in 3D space. This artifact is more prominent in basal and apical segments in which the position of the short-axis contours may change significantly because of through-plane motion. This could explain the poor correlation between rEF and SQUEEZ in apical and basal segments. In addition, rEF calculation is based on distance from a centerline: If the centerline is not chosen correctly, there will be large errors in rEF calculations. This artifact is more prominent in apical slices, where a small change in the position of the centerline could result in large rEF values. Although the centerlines were chosen carefully, the irregular shape of the endocardial contour in apical slices, especially in hearts that have undergone significant LV remodeling, would still cause very large rEF values in some data sets.
Comparison With Other Modalities
Echocardiography is widely used in clinics for cardiac function and dyssynchrony analysis but does not produce a detailed map of the coronaries and is limited by the available acoustic window. Although echocardiography has very high temporal resolution, the available window for transducer placement limits the orientation of the imaging plane. Furthermore, the variance of repeated echocardiographic measures is fairly high because of dependence on operator experience, machine settings, available acoustic windows, and angle of incidence effects.
CMR is another highly attractive modality for regional cardiac function analysis. However, this method also has some practical drawbacks, including higher cost, occasional gating failures, more-complex scanning protocols, and longer scanning times. The inability to study the growing population of patients with implanted electronic devices, which is especially relevant in patients with previous MI is another drawback. With its high isotropic resolution, CT decreases partial volume effects, accurately characterizes the blood-myocardium interface, and eliminates through-plane motion artifacts.
Radiation and Iodinated Contrast
Radiation exposure and iodinated contrast agent dose are still primary concerns with the increasing use of CT technology in diagnostic cardiac imaging and have become a centerpiece of new hardware, software, and imaging protocol improvements. Wide-range detector CT, with full cardiac coverage, does not need an overlapping radiation exposure and long scan times, as required in traditional helical CT imaging systems and, thus, provides a significant radiation and contrast dose savings for cardiac imaging.8,19 Significant leaps have been made in the dose reduction algorithms parallel to advances in hardware in the past couple of years. Furthermore, we extended our validations to tube current modulation of 10% and prospectively gated scans8 with tube currents as low as 5% of the high-dose scans, and the initial results were encouraging and confirmed the feasibility of using low-dose CT.
Iterative reconstruction algorithms20–22 could potentially improve the results of our method further; however, they were not available at the time of these measurements. When available, these algorithms and numerous CT image noise reduction methods23,24 are expected to improve the results of SQUEEZ by reducing the noise at current x-ray tube settings or reducing dose with tube current reduction. In the present study, we demonstrated that SQUEEZ is capable of quantifying regional cardiac function with routine CT acquisitions; determination of the minimum number of photons for clinically acceptable images requires further investigation.
Wide-range detector CT has limitations in temporal resolution intrinsic to CT imaging. This limitation has been reduced, with improvements in multibeat segmented image reconstruction and gantry rotation speed such that temporal resolution of <60 ms is now achievable.9 Although it is possible to calculate SQUEEZ from single-beat CT acquisitions, especially during systole, multibeat reconstructions will produce more-robust SQUEEZ values with higher temporal resolution.
The reliance on SQUEEZ alone, rather than strain measures, such as myocardial shortening, to characterize local function may be a limitation. Although SQUEEZ and myocardial strain both reflect mechanical contractile function of the heart, a simple mathematical relation between MRI midwall strain and SQUEEZ may not be found because they measure 2 different physical parameters. Strain measures shortening in the myocardium and is most reliable in the midwall because of partial volume artifacts near the epicardial and endocardial borders in MRI. SQUEEZ, on the other hand, reflects endocardial deformation by measuring local changes in the area of the endocardial surface. Endocardial deformation metrics like SQUEEZ have an advantage in detecting the myocardial ischemic cascade and the resulting transmural strain gradient.25 A recent study has shown that a significant correlation exists between surface deformation and 1D strain metrics in 3D speckle-tracking images.26 These results lead us to believe that there will be a linear relationship between SQUEEZ and strain metrics, but we have yet to determine the precise relationship between SQUEEZ and circumferential shortening under all circumstances. Although the data obtained in this study are promising, larger studies are needed to establish the precise diagnostic accuracy of SQUEEZ.
Sources of Funding
Funding for animal preparation was provided by National Institutes of Health grants R01-HL64795 and R01-HL094610 (Henry Halperin, principal investigator). A stipend for Mr Pourmorteza was partially provided by a Siemens PhD fellowship for imaging research.
Dr McVeigh has ownership interest in MRI Intervetions Inc. Dr McVeigh has intellectual property in the field (US Patent 6,171,241, January 9, 2001 for the method for measuring myocardial motion and the like, E. R. McVeigh and B. D. Bolster).
We thank Kristine Evers, Valeria Sena-Weltin, Jorge Guzman, and Theresa Caton for their assistance with animal preparation, CT acquisition, and data management.
- Received January 31, 2011.
- Accepted February 14, 2012.
- © 2012 American Heart Association, Inc.
- Helm RH,
- Leclercq C,
- Faris OP,
- Ozturk C,
- McVeigh E,
- Lardo AC,
- Kass DA
- Schuleri KH,
- Centola M,
- George RT,
- Amado LC,
- Evers KS,
- Kitagawa K,
- Vavere AL,
- Evers R,
- Hare JM,
- Cox C,
- McVeigh ER,
- Lima JAC,
- Lardo AC
- Cerqueira MD,
- Weissman NJ,
- Dilsizian V,
- Jacobs AK,
- Kaul S,
- Laskey WK,
- Pennell DJ,
- Rumberger JA,
- Ryan T,
- Verani MS
Determination of left ventricular regional function is important in the diagnosis and management of cardiomyopathy. We describe a novel, CT-based method, SQUEEZ (Stretch Quantifier for Endocardial Engraved Zones), that provides highly quantitative measures of regional myocardial function and can distinguish between infarcted and normally contracting myocardial regions with minimal user interaction and high resolution. This approach may have particular clinical value in assessing patients with dyssynchronous heart failure to better identify their candidacy for cardiac resynchronization therapy and may help to guide cardiac resynchronization therapy lead placement to the most appropriate myocardial location. Additionally, SQUEEZ may help to assess myocardial dysfunction in patients with myocardial ischemia, especially when used in tandem with CT coronary atherosclerosis and emerging CT regional blood flow assessment techniques.