First-Pass Contrast-Enhanced Myocardial Perfusion MRI Using a Maximum Up-Slope Parametric Map

574

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006

First-Pass Contrast-Enhanced Myocardial Perfusion MRI Using a Maximum Up-Slope Parametric Map Chun Ruan, Scott Yang, Geoffrey D. Clarke, Member, IEEE, Maxwell R. Amurao, Scott R. Partyka, Yong C. Bradley, and Kenneth Cusi

Abstract—Magnetic resonance first-pass perfusion imaging offers a noninvasive method for the rapid, accurate, and reproducible assessment of cardiac function without ionizing radiation. Quantitative or semiquantitative analysis of changes in signal intensity (SI) over the whole image sequence yields a more efficient analysis than direct visual inspection. In this paper, a method to generate maximum up-slope myocardial perfusion maps is presented. The maximum up-slope is defined by comparison of the SI variations using frame-to-frame analysis. A map of first-pass transit of the contrast agent is constructed pixel by pixel using a linear curve fitting model. The proposed method was evaluated using data from eight subjects. The data from the parametric maps agreed well with those obtained from traditional, manually derived region-ofinterest methods as shown through ANOVA. The straightforward implementation and increase in image analysis efficiency resulting from this method suggests that it may be useful for clinical practice. Index Terms—Magnetic resonance (MR) imaging, myocardial perfusion, parametric map, stress test.

I. INTRODUCTION YOCARDIAL perfusion is an important and sensitive indicator for the evaluation of patients with coronary artery disease because myocardial blood flow is directly correlated to oxygen supply [1]. Contrast-enhanced cardiac magnetic resonance (CMR) imaging provides an efficient measure of myocardial perfusion that is accurate, reproducible, and minimally invasive while utilizing nonionizing radiation [2], [3]. CMR also has superior temporal and spatial resolution in comparison with nuclear cardiologic imaging methods [4]. As such, CMR is being adopted as a clinical tool for myocardial perfusion imaging [5], [6].

M

Manuscript received January 15, 2004; revised November 25, 2004 and August 16, 2005. C. Ruan, G. D. Clarke, and M. R. Amurao are with the Department of Radiology, University of Texas Health Science Center at San Antonio, San Antonio, TX 78229 USA (e-mail: [email protected]; [email protected]; [email protected]). S. Yang was with the Department of Medicine, University of Texas Health Science Center at San Antonio, San Antonio, TX 78229 USA. He is now with the Department of Cardiology, Kaiser Permanente Santa Rosa, Santa Rosa, CA 95403 USA (e-mail: [email protected]). S. R. Partyka is with the South Texas Radiology Group, San Antonio, TX 78229 USA (e-mail: Scott partyka [email protected]). Y. C. Bradley is with the Department of Radiology, Brooke Army Medical Center, Fort Sam Houston, TX 78234 USA (e-mail: yong.bradley@us. army.mil). K. Cusi is with the Department of Medicine, University of Texas Health Science Center at San Antonio, San Antonio, TX 78229 USA (e-mail: cusi@ uthscsa.edu). Digital Object Identifier 10.1109/TITB.2006.872058

First-pass contrast-enhanced (FPCE) myocardial perfusion magnetic resonance imaging (MRI) measurements are able to identify small changes in myocardial blood flow. This capability is important for the detection of coronary artery disease [7]. FPCE myocardial perfusion MRI studies with fast multi-slice imaging sequences typically involve in obtaining 100–200 images for 2–5 slices of the heart (40–60 images per slice in a time series). These large volumes of data make visual inspection tedious and prone to observer bias [8], [9]. In addition, it is difficult to make comparisons between studies due to the lack of measurement standardization and quantitative information. As an alternative to visual evaluation, semiquantitative assessment of signal intensity (SI) changes over the course of the first-pass of a contrast agent through the myocardium offers a more efficient way for visualizing myocardial perfusion [10], [11]. Several quantitative or semiquantitative analysis techniques have been previously investigated. Among these, a linear fit of the SI(t) curve has been generally accepted as the most reliable parameter for evaluating myocardial perfusion in the clinic. The variation of the up-slope between myocardial sectors correlates well with differences in blood flow [3]. This fitted up-slope is also independent of the SI(t) curve washout characteristics, making it suitable for conditions in which peripheral injection of contrast agent is used. However, it is not possible to obtain quantitative flow data using this method. The conventional method used to generate the SI(t) curves from MR images is labor- and time-intensive because manual segmentation and calculation of the region-of-interest (ROI) areas from all the 100–200 images in the perfusion sequence is a tedious process [9]. Also, imaging parameter extraction and the subsequent up-slope calculation can be protracted. Usually, 4–6 ROI areas over the whole myocardium should be used for a thorough analysis. Assuming that there are 3–5 slices for each perfusion sequence, it may take up to 12 h to analyze all these data manually. Moreover, it is likely that performing this process by hand allows for significant human error since myocardial margins must be determined by scrolling through images in which first the right ventricular lumen enhances, then the left ventricular lumen stands out, and then the myocardium brightens. The lack of standardization of the manual method and duration of analysis reduces the utility of this technique for routine clinical measurements. In the present report, a method for automatically generating semiquantitative myocardial perfusion parametric maps depicting the maximum up-slope of the SI(t) curve is presented. The up-slope of each pixel over the heart can be calculated automatically in several seconds. Therefore, both the accuracy

1089-7771/$20.00 © 2006 IEEE

RUAN et al.: FPCE MYOCARDIAL PERFUSION MRI USING A MAXIMUM UP-SLOPE PARAMETRIC MAP

575

and efficiency of quantitative assessment may be improved compared to the manual ROI analysis method. Instead of requiring observation of all the perfusion images in a dataset, the perfusion information is reduced to a single parametric map per slice. In the following report, an algorithm is described that is designed to improve the diagnostic utility of FPCE myocardial perfusion MRI in a clinical setting. Analysis of the resulting data is used to test the hypothesis that the values obtained using this method show good correspondence to data obtained using the conventional manual method. II. METHODS Cardiac MRI data were transferred for analysis to a Dell workstation 400 (Dell Computer Corp., Round Rock, TX) and were processed using the algorithm, described below, which was written in IDL 5.6 (Research Systems, Inc., Boulder, CO). For each MR perfusion imaging sequence, the pixel values at each specific coordinate in successive images were extracted as one-dimensional (1-D) SI versus image time data array. A 1-D difference array was then produced by subtracting the neighboring elements of the 1-D SI data array, depicting the change in SI with time for each pixel. The largest element in the difference array implies that maximum contrast agent concentration variation occurs during this time interval. The largest element of the 1-D difference array and its corresponding element in the 1-D SI array were identified automatically. Only the positive maximum element in the difference array was selected, reflecting the first-pass up-slope parameter. Zero values were assigned to negative maximum up-slope values. Three consecutive data points in the 1-D SI array, centered about the identified element, were then extracted. The three data points represented the most rapid change in SI over time for the whole imaging sequence, at a particular coordinate. The three consecutive elements were fitted using a linear curvefitting model y = A + Bx. Errors in the fitting process were controlled by minimizing the Chi-square error statistic [12]. In this model, x is acquisition time, which is automatically obtained from the DICOM header file, y is SI, and B is the maximum SI up-slope during the contrast agent’s first-pass in the image series. The procedure was repeated for calculating the upslope of each pixel over the whole image under the assumption that each pixel value reflects the dynamic properties of contrast agent concentration in the myocardium. Thus, the image series is reduced to a single image (a parametric map) where each pixel value corresponds to the maximum up-slope of the SI(t) curve at that coordinate. Coordinates with the most rapid wash-in of contrast agent in the first-pass appear brightest, and regions with similar perfusion have equal gray-scale values. Hence, the maximum upslope parametric map can be used to quickly assess differences in myocardial perfusion. Setting a threshold for negative maximum up-slope values has a minimal effect on myocardial ROI areas because the negative up-slope values are typically found in the image background and adjacent organs, where little or no perfusion occurs. Another threshold can also be set for very high maximum up-slope values, which are typically observed in the

Fig. 1.

Flowchart of parametric map calculation.

ventricular cavity and are characteristic of rapid contrast agent wash-in. Suitable threshold selection can make the myocardium more pronounced in the generated maximum up-slope parametric map, which allows for straightforward identification of ischemic regions. The general scheme of the calculation of maximum up-slope parametric map is shown in Fig. 1. The Interactive Data Language (IDL) code can be made available by the authors upon request. Processed data were evaluated by linear regression with ANOVA methods set at a confidence level of 95% using GraphPad Prism (GraphPad Software, Inc., San Diego, CA). III. APPLICATION The algorithm was tested by using data from two studies acquired from different manufacturers’ systems, operating at different field strengths and using different pulse sequences. Study 1: CMR was performed using a 1.5-T MRI scanner (Edge Eclipse, Philips Medical Systems, Highland Heights, OH) at the Brooke Army Medical Center (BAMC) at Fort Sam Houston, TX. Myocardial FPCE data, obtained from six diabetic

576

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006

subjects suspected of having heart disease (four male, two female) with ages ranging from 40 to 75 years (mean = 65 years) were used to evaluate the accuracy of the algorithm. Three of the subjects were diagnosed with a myocardial perfusion defect from both Nuclear Cardiologic and CMR techniques. The remaining three subjects were diagnosed as defect-free under both imaging modalities. These studies were approved by the BAMC Institutional Review Board (IRB), and patient information was preserved in accordance with the Health Insurance Portability and Accountability Act (HIPAA) standards. A phased array cardiac coil was used with the magnetization driven equilibrium Fourier transform (MDEFT) sequence with the following parameters: TE = 1.5 ms, θ = 40◦ , slice thickness = 1.0 cm, TR = 2 R–R intervals with electrocardiogram-triggered cardiac gating (ECG) and breath-hold respiratory motion compensation [13]. Study 2: Perfusion data were obtained from the 3.0-T MRI system (MAGNETOM Trio, Siemens Medical Solutions USA, Inc., Malvern, PA) at the Research Imaging Center of the University of Texas Health Science Center at San Antonio, TX, from two normal subjects (one male and one female, ages 40– 48 years). These studies were approved by the UT Health Science Center IRB. An eight-channel phased array cardiac radio frequency coil (USA Instruments, Aurora, OH) was used in conjunction with parallel imaging to accelerate the acquisition by a factor of two. A Turbo-FLASH saturation recovery sequence was used with a 90◦ saturation pulse followed by a series of excitation pulses acquiring 60 phase-encode acquisitions per image, TR/TE = 188/0.93 ms, θ = 18◦ , slice thickness = 6 mm. ECG triggering and breath-holding methods are controlled for motion artifacts [14]. For both studies, two data sets were collected from each subject when lying supine under normal resting conditions (rest state) and during the intravenous (IV) infusion of adenosine (pharmacologic stress). After 4 min of adenosine infusion (Fujisawa 140 µg/kg/min) using an MRI-compatible infusion system (Continuum, Medrad Inc., Pittsburgh, PA), a dose of a contrast agent containing gadolinium, Gadoversetamide (OptiMARK 0.05 mmol/kg, Tyco Healthcare, Mansfield MA) was injected using an MRI-compatible power injector (Spectris, MEDRAD Inc., Pittsburgh, PA). Short-axis images were acquired during and immediately following the 3 cc/s gadolinium injection. These FPCE myocardial perfusion MR images were collected under pharmacologically induced hyperemia. Ten to twenty minutes following cessation of adenosine infusion, a repeat dynamic contrast-enhanced MRI was performed with gadolinium without adenosine infusion to acquire data under a baseline resting condition. For the first study (1.5-T MRI system), the second collection of images represented the FPCE myocardial perfusion MRI at rest. For the second study, conducted on the 3.0-T MRI system, the order of acquisition was reversed. The timing parameters of the cardiac MRI perfusion acquisition schemes were set to allow one set of image slices to be obtained. The algorithm described above was employed and parametric maps depicting the up-slope of the uptake curve were generated for each image slice. If the patient’s heart rate changed upon administration of adenosine, then the operator

Fig. 2. (a)–(d) Successive MRI perfusion images at rest of a patient with myocardial perfusion defect in the lateral wall of the left ventricle. (e), (f) Maximum up-slope parametric map without and with upper threshold, respectively, which were generated from the same image sequence at the same slice as (a)–(d).

Fig. 3. Maximum up-slope of signal-intensity curve extracted from three consecutive data points using a 36-pixel square ROI within the septal and lateral walls of the left-ventricular myocardium, respectively. Each data point is from a separate image acquired every 2 R–R intervals.

would adjust the number of phase encoding steps to ensure that an image set was acquired for each heartbeat or every other heartbeat. Since the image acquisition time (150–300 ms) is short compared to the rate of uptake of the contrast agent in the myocardium, the algorithm that produces the parametric map was not adjusted for changes in heart rate. The parametric maps depicting the up-slope of the uptake curve were generated for each image slice. A sample image from the 1.5-T MRI scanner and its corresponding parametric map are shown in Fig. 2. For the data from Study 1 (1.5 T), a comparison between the proposed algorithm and the traditional ROI method was performed. A 36-pixel square ROI was centered within the septal, lateral, anterior, and posterior walls of the image of the left ventricular myocardium. The ROI areas were prescribed for each image slice to include exactly the same pixels. The mean SI within the ROI in each image was measured for all images in the time series for each slice, and the SI(t) uptake curve was generated. The maximum up-slope of the uptake curve for each ROI was extracted from the three consecutive data points that exhibited the highest increases in SI (Fig. 3). With four ROI areas per slice, three slices per series, baseline and pharmacologically induced hyperemic conditions for six subjects, a total of 144 sets of up-slope values were obtained. Data from the 36-pixel square ROI areas in the parametric map corresponding to the ROI areas prescribed in the ROI method were extracted, and the parametric map’s mean pixel value (up-slope)

RUAN et al.: FPCE MYOCARDIAL PERFUSION MRI USING A MAXIMUM UP-SLOPE PARAMETRIC MAP

577

ated using the mean value from each segment ROI. The up-slope of the uptake curve was calculated by taking the average of the three greatest consecutive differences between subsequent ROI areas. For comparison, the calculated parametric map was segmented in the same manner and the up-slope was calculated using the mean signal intensities from the ROI segments. ROI areas were prescribed for each of the three slices, with each subject scanned during both rest and adenosine stress. Thus, 36 measurements were performed for each subject. Although the size of the segment ROI areas varied with subject and image slice, the average segment ROI area was 147.7 cm2 (SD = 53.2 cm2 ). Again a small but significantly lower value of the upslope of the SI(t) curve was measured using the traditional ROI method compared with the value obtained from the parametric map (∆ = −0.94, range = 2.98–10.90, p  0.001). Linear regression of the data showed a significant correlation (r = 0.87; slope = 0.93, p  0.0001).

Fig. 4. Scatter plot comparing calculated values for the maximum up-slope in the SI(t) curve for study #1 data (from the 1.5-T MRI system) shows that the data from the parametric map algorithm significantly correlated with the ROI data (r = 0.99; p  0.0001).

was calculated for each ROI. A slightly smaller but significantly different value of the up-slope of the SI(t) curve was measured using the traditional ROI method compared with the value obtained from the parametric map using Student’s t-test, paired for two samples (∆ = −0.12, range = 0.86–27.44, p = 0.014). Linear regression of the data revealed a strong correlation (r = 0.99; slope = 1.04, p  0.0001) as depicted in Fig. 4. The standard deviations of the two distributions were similar (4.28 for pixel-by-pixel and 4.48 by ROI) suggesting that the noise levels of the measurements were also similar. By taking the up-slope value, determined during pharmacologic stress, and dividing it by the corresponding up-slope determined at baseline for each myocardial segment of each individual, a myocardial perfusion reserve index (MPRI) could be determined that was related to the functional capacity of the myocardium [6], [15]. These data showed a small but significant decrease (∆ = −0.15, range = 0.19–6.53, p = 0.014) in MPRI using the ROI method compared to the parametric map data. There was a weaker but still highly significant correlation between the MPRI values derived from the two methods (r = 0.93, slope = 1.12, p  0.001). The second test of the algorithm was more closely related to a true clinical evaluation of perfusion. The image of the left ventricular myocardium exhibiting the greatest contrast between the ventricular lumen and the myocardium was selected using a DICOM-compatible cardiac MRI viewing and analysis application (OSIRIS, Hospital of the University of Geneva, Switzerland). The myocardium was divided into six ROI areas using a standard segmentation scheme [9], [11]. The segmented ROI areas were then propagated to each of the images in the time series. The images in the time series were reviewed and ROI areas displaced from the myocardium were moved back over the myocardium. A time–intensity curve, SI(t), was gener-

IV. DISCUSSION A maximum up-slope calculated parametric map algorithm has been described and compared with the manual ROI method using the same ROI for eight patients. The proposed method is significantly more efficient in that the up-slope values are automatically computed on a pixel-by-pixel basis and then only one manually drawn ROI is required per myocardial segment. This reduces the amount of time spent on manual processing by an order of magnitude. The data show good correspondence between this standard manual evaluation method and the parametric map method for images from two MRI systems at different field strengths. A strong correlation was demonstrated between the data obtained using the two methods and a small but significant difference in the up-slopes was found, which suggests a systematic difference in the measurements resolvable by simple calibration methods. These results suggest that the maximum up-slopes calculated using the parametric map algorithm, as presented in this study, are comparable to those obtained manually. This algorithm is also fast and reproducible in a clinical setting. The average computer plus manual processing time for each data set using our algorithm is less than 1 h compared to 4–5 h using ROI method with a good user interface. Unless there is significant patient motion during the uptake of contrast agent, the generated parametric map will have high spatial resolution since the maximum up-slope is calculated for each pixel, making results more sensitive to small ischemic regions. To obtain better visualization of the myocardium, two threshold values were set for minimizing the contributions of the background and ventricle, respectively. Threshold selection is a critical issue for this process. In this paper, the threshold values were chosen by a standard histogram-based method. This resulted in adequate separation of the ventricular signal from background noise. However, asymmetric subendocardial margins can introduce bias in the pixels which straddle this skewed boundary. Typically, these pixels have averaged together the two principal brightness levels in proportion to the area subtended within the pixel, so that the boundary position becomes somewhat uncertain, as seen in Fig. 2 [23], [24]. Other robust image

578

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006

segmentation methods could be used in future studies to achieve higher levels of accuracy at the cost of much computation time. Nagel et al. have previously described the manual process of dividing the myocardium into six equiangular segments per slice [9]. The left ventricular SI of the most basal slice and myocardial SI were determined at all time points. Mean SI values before contrast agent injection were subtracted from all post-contrast values and the up-slopes of the resulting SI(t) curves were determined by using a linear fit of the data from four consecutive images. For each perfusion MR imaging sequence the mean SI(t) curve was generated in a user-defined ROI within the subject’s myocardium. The position of the ROI was then adjusted to match the same anatomical region in a series of images. The ROI sizes and shapes were kept constant throughout the analysis of each myocardial section. Baseline SI, before contrast enhancement, was subtracted from the SI(t) curve prior to image analysis. ROI-specific SI(t) curves were then analyzed with parameters such as the up-slope, down-slope, maximal SI, and the time to maximal SI being evaluated. Other investigators have proposed improvements to analyze CMR perfusion data. Jerosch-Herold et al. reported a pixel-bypixel analysis method for the signal time course [8]. The SI(t) curve for each pixel was smoothed by computing a least-square constrained spline approximation. The discussion principally concerned their elastic-matching algorithm, which compensated for subject motion. Details and validation of their parametric upslope algorithm were lacking. Also these investigators did not present any data acquired with their canine subjects experiencing pharmacologic stress. Germain et al. derived input and output functions using a complex algorithm from the time-intensity curves obtained in the left ventricular and myocardial ROI areas with 200 iterations. The algorithm was applied to each pixel and provided parametric images in terms of the tissue extraction coefficient, the flow per unit mass, and the tissue blood partition coefficient [16]. Although this method is elegant, it relies on multiple assumptions of the underlying physiology that would be difficult to validate in a clinical setting. Sipola et al. drew regions for the same 12 tissue segments manually and all of the SI(t) curves were fitted with the extended Freundlich model for a right-skewed curve. The maximal SI increase was determined from the fitted curves. The SI change rate was calculated as SI increase versus time [17]. Schwitter et al. used a sliding window to extract the maximal up-slope in each pixel of the ROI drawn manually [18]. All these methods use pixel-by-pixel analysis with different algorithms that eventually lead to a corresponding parametric image. However, they all include some manual intervention such as selection of an ROI, etc. [19], [20]. None of the investigators, listed above, have provided details on the implementation of their algorithms. Specifically missing, and of some importance, are methods used to define the starting and ending point of the up-slope calculation [21]. In the current algorithm, the three data points can be extracted automatically without any requirement for manual selection of starting and ending images as has been indicated for previously published algorithms. The reliability of the curve fitting using the current algorithm is the most important property that was tested. Also, the

purpose of the work is to make progress toward a simple and practical way to quantitatively evaluate the myocardial perfusion of the patient in clinical setting. The manual ROI method is the most widely used technique in clinical studies and therefore was selected for comparison. The results revealed a small but significant difference between both the up-slope of SI(t) and the MRPI results, which suggests that direct comparisons between the two methods will require calibration of the respective data sets. We note that the values of the up-slopes obtained in the current study from these two MRI systems were very different since the scaling of SI units is not absolute, but relative to the receiver electronics and bit depth of the digitizer. Thus, more direct comparisons of perfusion indices obtained on such different MRI systems will require calibration to a common standard. There is currently no method incorporated to account for patient motion and therefore the proposed method may be rendered useless if the experimental subject cannot comply with the requirement for breath holding. One approach to minimize this source of error would be to evaluate perfusion in only the central regions of myocardium, as was done in study #1 of the present project; however, this would detract from the potential ability of cardiac MRI to evaluate transmural perfusion. Improvements in protocol implementation may allow data for the up-slope of the SI(t) uptake curve to be collected before breath-hold failure, even for patients with compromised respiration, since only the data points in three continuous images are used for the final up-slope calculation. If necessary, motion correction should be performed prior to further analysis. However, this topic is beyond the scope of the current paper. More advanced approaches have been proposed for a fully quantitative assessment of myocardial perfusion [22]. Quantitative analysis is based on the absolute myocardial blood flow and has been derived from the hemodilution theory. Additional requirements must be met for reliable quantitative analysis: a nondiffusible tracer, complete wash-out of the tracer, a single spike input function, and a linear correlation between the tracer and the SI [3]. However, it is hard to fulfill these requirements for the current MR perfusion study using diffusible contrast agents. The parameters obtained from these models describe not only myocardial perfusion, but also the distribution volume of the contrast agent within myocardium. Although these approaches have been validated in animal experiments, the analyses are conceptually complicated, time consuming, operator dependent, and require a large number of assumptions. The improved efficiency reduced manual intervention and reproducibility of this algorithm suggests its suitability for clinical comparisons. The immediate next step in this research is to employ this algorithm to produce maps of the perfusion reserve index, obtained by a division of the up-slope measured under stress conditions to the up-slope obtained under baseline conditions. However, slight changes in the patient position and changes in the heart rate in the 10–20 min between scans can move the sections of the heart that are imaged. Thus, a method employing between slice image interpolation and/or image warping will be required to make robust perfusion reserve maps, which is the focus of current investigations.

RUAN et al.: FPCE MYOCARDIAL PERFUSION MRI USING A MAXIMUM UP-SLOPE PARAMETRIC MAP

V. CONCLUSION A new method has been proposed for the automated generation of maximum up-slope parametric maps for a semiquantitative description of FPCE myocardial perfusion MRI data. Serial images, obtained from eight subjects were analyzed and calculated values for maximum up-slope of the SI(t) uptake curve were found to be comparable with the traditional manual ROI method. These data suggest that maximum up-slope parametric maps, using the algorithm presented in this study, may provide a useful method for rapid, objective evaluation of clinical FPCE myocardial perfusion MRI studies. REFERENCES [1] N. Al-Saadi et al., “Noninvasive detection of myocardial ischemia from perfusion reserve based on cardiovascular magnetic resonance,” Circulation, vol. 101, pp. 1379–1383, 2000. [2] O. Muhling, M. Jerosch-Herold, M. Nabauer, and N. Wilke, “Assessment of ischemic heart disease using magnetic resonance first-pass perfusion imaging,” Herz, vol. 28, pp. 82–89, 2003. [3] E. Nagel, N. Al-Saadi, and E. Fleck, “Cardiovascular magnetic resonance: Myocardial perfusion,” Herz, vol. 25, pp. 409–416, 2000. [4] J. R. Panting, P. D. Gatehouse, G. Z. Yang, M. Jerosch-Herold, N. Wilke, D. N. Firmin, and D. J. Pennell, “Echo-planar magnetic resonance myocardial perfusion imaging: Parametric map analysis and comparison with thallium SPECT,” J. Magn. Reson. Imag., vol. 13, pp. 192–200, 2001. [5] T. Laddis, W. J. Manning, and P. G. Danias, “Cardiac MRI for assessment of myocardial perfusion: Current status and future perspectives,” J. Nucl. Cardiol., vol. 8, pp. 207–214, 2001. [6] P. R. Sensky et al., “Coronary artery disease: Combined stress MR imaging protocol—one-stop evaluation of myocardial perfusion and function,” Radiology, vol. 215, pp. 608–614, 2000. [7] A. Zenovich, O. M. Muehling, P. M. Panse, M. Jerosch-Herold, and N. Wilke, “Magnetic resonance first-pass perfusion imaging: Overview and perspectives,” Rays, vol. 26, pp. 53–60, 2001. [8] M. Jerosch-Herold and N. Wilke, “MR first pass imaging: Quantitative assessment of transmural perfusion and collateral flow,” Int. J. Card. Imag., vol. 13, pp. 205–218, 1997. [9] E. Nagel, C. Klein, I. Paetsch, S. Hettwer, B. Schnackenburg, K. Wegscheider, and E. Fleck, “Magnetic resonance perfusion measurements for the noninvasive detection of coronary artery disease,” Circulation, vol. 108, pp. 432–437, 2003. [10] H. Penzkofer, B. J. Wintersperger, A. Knez, J. Weber, and M. Reiser, “Assessment of myocardial perfusion using multisection first-pass MRI and color-coded parameter maps: A comparison to 99mTc Sesta MIBI SPECT and systolic myocardial wall thickening analysis,” Magn. Reson. Imag., vol. 17, pp. 161–170, 1999. [11] A. O. Boudraa, F. Behloul, M. Janier, E. Canet, J. Champier, J. P. Roux, and D. Revel, “Temporal covariance analysis of first-pass contrast-enhanced myocardial magnetic resonance images,” Comput. Biol. Med., vol. 31, pp. 133–142, 2001. [12] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. F. Flannery, Numerical Recipes in C ++ : The Art of Scientific Computing, 2nd ed. Cambridge, UK: Cambridge Univ. Press, 2002, ch. 15. [13] N. V. Tsekos, Y. Zhang, H. Merkle, N. Wilke, M. Jerosch-Herold, A. Stillman, and K. Ugurbil, “Fast anatomical imaging of the heart and assessment of myocardial perfusion with arrhythmia insensitive magnetization preparation,” Magn. Reson. Med., vol. 34, pp. 530–536, 1995. [14] J. Frahm and A. Hasse, “Rapid NMR imaging of dynamic processes using the FLASH technique,” Magn. Reson. Med., vol. 3, pp. 321–327, 1986. [15] S. Plein et al., “Coronary artery disease: Assessment with a comprehensive mr imaging protocol—initial results,” Radiology, vol. 225, pp. 300–307, 2002. [16] P. Germain et al., “Myocardial flow reserve parametric map, assessed by first-pass MRI compartmental analysis at the chronic stage of infarction,” J. Magn. Reson. Imag., vol. 13, pp. 352–360, 2001. [17] P. Sipola et al., “First-pass MR imaging in the assessment of perfusion impairment in patients with hypertrophic cardiomyopathy and the Asp175Asn mutation of the alpha-tropomyosin gene,” Radiology, vol. 226, pp. 129–137, 2003.

579

[18] J. Schwitter et al., “Assessment of myocardial perfusion in coronary artery disease by magnetic resonance: A comparison with positron emission tomography and coronary angiography,” Circulation, vol. 103, pp. 2230– 2235, 2001. [19] T. Ibrahim, S. G. Nekolla, K. Schreiber, K. Odaka, S. Volz, and J. Mehilli, “Assessment of coronary flow reserve: Comparison between contrastenhanced magnetic resonance imaging and positron emission tomography,” J. Amer. Coll. Cardiol., vol. 39, pp. 864–870, 2002. [20] F. H. Epstein, J. F. London, D. C. Peters, L. M. Goncalves, K. Agyeman, and J. Taylor, “Multislice first-pass cardiac perfusion MRI: Validation in a model of myocardial infarction,” Magn. Reson. Med., vol. 47, pp. 482– 491, 2002. [21] K. M. Bertschinger, D. Nanz, M. Buechi, T. F. Luescher, B. Marincek, G. K. von Schulthess, and J. Schwitter, “Magnetic resonance myocardial first-pass perfusion imaging: Parameter optimization for signal response and cardiac coverage,” J. Magn. Reson. Imag., vol. 14, pp. 556–562, 2001. [22] J. Barkhausen, P. Hunold, M. Jochims, and J. F. Debatin, “Imaging of myocardial perfusion with magnetic resonance,” J. Magn. Reson. Imag., vol. 19, pp. 750–757, 2004. [23] J. C. Russ, The Image Processing Handbook, 2nd ed. Boca Raton, FL: CRC Press, pp. 376–379. 1995. [24] K. R. Castleman, Digital Image Processing. Englewood Cliffs, NJ: Prentice-Hall, 1996, pp. 460–462.

Chun Ruan received the Master’s degree in medical imaging from Taishan Medical College, Tai’an, China, in 2002, and the Ph.D. degree in radiological sciences from the University of Texas Health Science Center, San Antonio, in 2005. She is currently a Postdoctoral Fellow at the University of Texas Health Science Center. Her research interests include development of cardiac MRI for the study of coronary artery disease and medical image processing.

Scott Yang received the undergraduate and Ph.D. degrees from Harvard University, Cambridge, MA, both in physics. After completing the Ph.D. degree, he went to Harvard Medical School and completed his cardiology fellowship at the University of Texas Health Science Center, San Antonio, in 2005. He is currently a Cardiologist at Kaiser Permanente Santa Rosa, Santa Rosa, CA.

Geoffrey D. Clarke (S’84–M’84) received the Ph.D. degree in radiological sciences from the University of Texas Southwestern Medical Center, Dallas, in 1984. He is a Professor and Vice-Chair of Radiology and Director of the Graduate Program in Radiological Sciences at the University of Texas Health Science Center, San Antonio. His research interests include cardiovascular imaging and image quality testing methodologies.

580

IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 10, NO. 3, JULY 2006

Maxwell R. Amurao received the Bachelor’s degree in applied physics from the University of the Philippines, Manila, Philippines, in 1993, and the Master’s degree (with distinction) in physics from De La Salle University, Manila, Philippines, in 2002. He is currently working toward the Ph.D. degree in radiological sciences at the University of Texas Health Science Center, San Antonio. He is also a Medical Physicist with the Baylor Health Care System in the Dallas-Fort Worth, TX, area, currently involved with instruction and clinical applications of diagnostic radiologic physics, medical nuclear physics, and radiation safety.

Scott R. Partyka, photograph and biography not available at the time of publication.

Yong C. Bradley, photograph and biography not available at the time of publication.

Kenneth Cusi received the M.D. degree and completed his residency in Buenos Aires, Argentina. He completed a fellowship in endocrinology, diabetes, and metabolism at Baylor College of Medicine, Houstan, TX. He is an Associate Professor in the Diabetes Division, University of Texas Health Science Center, San Antonio, and the Director of the Endocrinology and Metabolism Clinic, Veterans Administration Medical Center, San Antonio, TX. His research interests include understanding the mechanisms of insulin resistance and the associated abnormalities that cause cardiovascular disease, as well as searching for new pharmacological treatments for patients with type 2 diabetes.