HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text Full Text (PDF) Appendix E1 Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Wen, H. Articles by Croisille, P. Search for Related Content PubMed PubMed Citation Articles by Wen, H. Articles by Croisille, P.

Adaptive Postprocessing Techniques for Myocardial Tissue Tracking with Displacement-encoded MR Imaging¹

¹ From the National Heart, Lung and Blood Institute, National Institutes of Health, Bldg 10, B1D416, 10 Center Dr, Bethesda, MD 20892 (H.W., E.E.B., R.P.L., N.D.E.); Department of Computer Science and Engineering, Ohio State University, Columbus, Ohio (K.A.M.); Division of Cardiology, George Washington University School of Medicine, Washington, DC (J.F.P.); Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY (K.S.K.); Department of Biomedical Engineering, Tulane University, New Orleans, La (D.B.L.); and Department of Radiology, Hôpital Cardiologique et Pneumologique, L. Pradel, Lyon, France (P.C.). Received January 9, 2007; revision requested March 2; revision received March 28; accepted May 2; final version accepted July 2. Supported by the National Heart, Lung, and Blood Institute, National Institutes of Health. Address correspondence to H.W. (e-mail: wenh{at}nhlbi.nih.gov).

View larger version (8K): [in this window] [in a new window] [Download PPT slide]   Figure 1: Flow diagram shows patient inclusion and exclusion. Of the 17 consecutive eligible patients, one patient was excluded because electrocardiographic (ECG) signal was insufficient to trigger the imager. The remaining 16 patients underwent displacement-encoded cardiac MR imaging and two-dimensional (2D) echocardiography. Two-dimensional echocardiography revealed abnormal myocardial wall motion—including hypokinetic, dyskinetic, and akinetic segments—in 11 patients.

View larger version (12K): [in this window] [in a new window] [Download PPT slide]   Figure 2: The pulse sequence of displacement-encoded cardiac MR imaging was triggered by the R wave of the electrocardiographic (ECG) signal. The position-encoding section included the first two 90° pulses and the encoding gradient pulses between them. Position-encoding and position-decoding gradient pulses are shaded in this diagram. Readout portion is a two-dimensional cine DENSE sequence with short-echo-train echo-planar imaging readout. It is used to measure three-dimensional displacement vectors by acquiring four cine data sets in different encoding directions. These are as follows: (a) Y, Z; (b) –Y, Z; (c) X, Z; and (d) X, –Z, where X, Y, and Z are the readout, phase-encode, and section directions, respectively. Decoding gradient in the phase-encode direction is merged with phase-encode gradient; this results in the three types of phase-encode gradient pulses shown. The echo train length is six per radiofrequency (RF) excitation (schematically shown as two), and in each cardiac phase, 24 k-space lines were acquired in four radiofrequency excitations that lasted 31 msec. This was repeated (dotted lines) for 17–20 cardiac phases, depending on the heart rate. The image matrix was 128 x 48, and two signals were acquired. A total of 16 heartbeats were needed for the complete data set.

View larger version (15K): [in this window] [in a new window] [Download PPT slide]   Figure 3: Outline of the postprocessing procedure. Ecc = circumferential strain, LV = left ventricle, RV = right ventricle.

View larger version (115K): [in this window] [in a new window] [Download PPT slide]   Figure 4: Three sets of left and right ventricular contours of the last (20th), 15th, and 13th cine frames are shown on the last cine frame of the midleft ventricle short-axis view. The contours of the intermediate frames have been omitted for clarity. The outermost contours are of the last frame and have been drawn manually. These are automatically updated for the other cine frames on the basis of the displacement vectors. Arrows indicate the sequence of contour generation. They point from contour positions in a later cine frame to those in an earlier frame in the reverse tracking process (Appendix E1, http://radiology.rsnajnls.org/cgi/content/full/246/1/229/DC1).

View larger version (66K): [in this window] [in a new window] [Download PPT slide]   Figure 5a: (a) The kernel of a fixed filter is identical at all locations. This short-axis view shows identical circular kernels at two locations in the left ventricle. (b) The kernel of an adaptive filter is always aligned with the local circumferential direction. The same short-axis view shows the arc-shaped kernels at two locations in the left ventricle.

View larger version (66K): [in this window] [in a new window] [Download PPT slide]   Figure 5b: (a) The kernel of a fixed filter is identical at all locations. This short-axis view shows identical circular kernels at two locations in the left ventricle. (b) The kernel of an adaptive filter is always aligned with the local circumferential direction. The same short-axis view shows the arc-shaped kernels at two locations in the left ventricle.

View larger version (41K): [in this window] [in a new window] [Download PPT slide]   Figure 6a: (a) Magnitude image and initial X, Y, and Z displacement-encoded phase maps. (b, c) Illustrations of phase unwrapping of an end-systolic cine frame of a short-axis view obtained in a volunteer. Spatial resolution is 3.0 mm. In-plane projections of three-dimensional displacement vectors are shown. The X-, Y-, and Z-encoded phase maps are shown at the top of each image. In b, the conventional phase-unwrapping technique leaves a phase fringe (arrows) in the anterior right ventricle. In c, the APU technique removes errors and pushes phase fringes beyond ventricular walls.

View larger version (61K): [in this window] [in a new window] [Download PPT slide]   Figure 6b: (a) Magnitude image and initial X, Y, and Z displacement-encoded phase maps. (b, c) Illustrations of phase unwrapping of an end-systolic cine frame of a short-axis view obtained in a volunteer. Spatial resolution is 3.0 mm. In-plane projections of three-dimensional displacement vectors are shown. The X-, Y-, and Z-encoded phase maps are shown at the top of each image. In b, the conventional phase-unwrapping technique leaves a phase fringe (arrows) in the anterior right ventricle. In c, the APU technique removes errors and pushes phase fringes beyond ventricular walls.

View larger version (64K): [in this window] [in a new window] [Download PPT slide]   Figure 6c: (a) Magnitude image and initial X, Y, and Z displacement-encoded phase maps. (b, c) Illustrations of phase unwrapping of an end-systolic cine frame of a short-axis view obtained in a volunteer. Spatial resolution is 3.0 mm. In-plane projections of three-dimensional displacement vectors are shown. The X-, Y-, and Z-encoded phase maps are shown at the top of each image. In b, the conventional phase-unwrapping technique leaves a phase fringe (arrows) in the anterior right ventricle. In c, the APU technique removes errors and pushes phase fringes beyond ventricular walls.

View larger version (15K): [in this window] [in a new window] [Download PPT slide]   Figure 7: Graph shows mean failure rates and 95% confidence intervals (error bars) of the conventional phase-unwrapping (white bars) and APU (black bars) techniques for a range of threshold of maximum phase gradient T(PGmax) values. Two-way analysis of variance was used to detect differences between the two techniques. Threshold of maximum phase gradient and technique served as the two factors. The APU technique had lower failure rates than did the conventional phase-unwrapping technique (P < .001). We compared the lowest failure rates of the two techniques and found that APU enabled reduction of the failure rate from 18.9% to 0.60% (P < .001).

View larger version (11K): [in this window] [in a new window] [Download PPT slide]   Figure 8a: Graphs show the effect of filters on noise levels, and systematic biases of strain maps are shown for three filter lengths. Mean and 95% confidence intervals (error bars) are shown for (a, b) Ecc and (c, d) Err. Filter types were compared with two-way analysis of variance, with filter length and type as the two factors. Adaptive filters (black bars) led to lower noise levels (a) (P < .001) and biases (b) (P < .001) in Ecc than did fixed filters (white bars). Adaptive filters also led to lower noise levels (c) (P < .001) and biases (d) (P = .029) in Err.

View larger version (10K): [in this window] [in a new window] [Download PPT slide]   Figure 8b: Graphs show the effect of filters on noise levels, and systematic biases of strain maps are shown for three filter lengths. Mean and 95% confidence intervals (error bars) are shown for (a, b) Ecc and (c, d) Err. Filter types were compared with two-way analysis of variance, with filter length and type as the two factors. Adaptive filters (black bars) led to lower noise levels (a) (P < .001) and biases (b) (P < .001) in Ecc than did fixed filters (white bars). Adaptive filters also led to lower noise levels (c) (P < .001) and biases (d) (P = .029) in Err.

View larger version (11K): [in this window] [in a new window] [Download PPT slide]   Figure 8c: Graphs show the effect of filters on noise levels, and systematic biases of strain maps are shown for three filter lengths. Mean and 95% confidence intervals (error bars) are shown for (a, b) Ecc and (c, d) Err. Filter types were compared with two-way analysis of variance, with filter length and type as the two factors. Adaptive filters (black bars) led to lower noise levels (a) (P < .001) and biases (b) (P < .001) in Ecc than did fixed filters (white bars). Adaptive filters also led to lower noise levels (c) (P < .001) and biases (d) (P = .029) in Err.

View larger version (10K): [in this window] [in a new window] [Download PPT slide]   Figure 8d: Graphs show the effect of filters on noise levels, and systematic biases of strain maps are shown for three filter lengths. Mean and 95% confidence intervals (error bars) are shown for (a, b) Ecc and (c, d) Err. Filter types were compared with two-way analysis of variance, with filter length and type as the two factors. Adaptive filters (black bars) led to lower noise levels (a) (P < .001) and biases (b) (P < .001) in Ecc than did fixed filters (white bars). Adaptive filters also led to lower noise levels (c) (P < .001) and biases (d) (P = .029) in Err.

View larger version (30K): [in this window] [in a new window] [Download PPT slide]   Figure 9: Bland-Altman plot shows pixel-wise differences between ASF and high SNR reference standard maps of Ecc for a filter length of 3 pixels. Means and standard deviations (STD) of the differences on such plots were used to indicate systematic biases and noise levels in filtered data.

View larger version (20K): [in this window] [in a new window] [Download PPT slide]   Figure 10: Average segmental Ecc curves of the left and right ventricles for all volunteers. Solid lines are the mean, and dashed lines are 1 standard deviation. The abscissa is the time after the electrocardiographic R wave (measured in milliseconds). Short-axis views at three levels are included. Temporal resolution was 31 msec, and 20 cine frames were acquired.

View larger version (15K): [in this window] [in a new window] [Download PPT slide]   Figure 11: Variability of segmental strain values among healthy volunteers was found to be lower with an ASF than with a fixed filter. P values were less than .001 (*) and equal to .020 (**).

View larger version (99K): [in this window] [in a new window] [Download PPT slide]   Figure 12a: The effect of adaptive versus fixed filters on the Ecc distribution is shown in midlevel short-axis views obtained in a patient with myocardial infarction. Baseline Ecc maps obtained (a) without filters, (b) with a 3-pixel fixed filter, and (c) with a 3-pixel ASF. Different results from adaptive and conventional spatial filters led to different sensitivities in the detection of abnormal segments in patients compared with two-dimensional echocardiographic examination results.

View larger version (98K): [in this window] [in a new window] [Download PPT slide]   Figure 12b: The effect of adaptive versus fixed filters on the Ecc distribution is shown in midlevel short-axis views obtained in a patient with myocardial infarction. Baseline Ecc maps obtained (a) without filters, (b) with a 3-pixel fixed filter, and (c) with a 3-pixel ASF. Different results from adaptive and conventional spatial filters led to different sensitivities in the detection of abnormal segments in patients compared with two-dimensional echocardiographic examination results.

View larger version (98K): [in this window] [in a new window] [Download PPT slide]   Figure 12c: The effect of adaptive versus fixed filters on the Ecc distribution is shown in midlevel short-axis views obtained in a patient with myocardial infarction. Baseline Ecc maps obtained (a) without filters, (b) with a 3-pixel fixed filter, and (c) with a 3-pixel ASF. Different results from adaptive and conventional spatial filters led to different sensitivities in the detection of abnormal segments in patients compared with two-dimensional echocardiographic examination results.

View larger version (9K): [in this window] [in a new window] [Download PPT slide]   Figure 13: ROC curves of MR imaging classification of normal (MR negative) and abnormal (MR positive) left ventriclular segments relative to echocardiographic normal (negative echocardiographic findings) and abnormal (positive echocardiographic findings) classification (reference standard) for three postprocessing configurations with conventional techniques, with APU only, and with both APU and ASF. MR classification was based on comparison of segmental Ecc with a threshold value. The range of Ecc thresholds that produced the ROC curves was –0.3 to 0.0 in increments of 0.0005.