- Email: [email protected]
Automatic left ventricular volume measurements on contrast-enhanced ultrafast cine magnetic resonance imaging
EUROPEAN JOURNAL OF RADIOLOGY
ELSEVIER
European Journal of Radiology 20 (1995) 126-132
Automatic left ventricular volume measurements on contrast-enhanced ultrafast cine magnetic resonance imaging Kentaro Matsumura *a, Emiko Nakase a, Tohru Haiyama b, Shinichi Utsunomiya c aDepartment of Internal Medicine, Kyoto Minami Hospital, 8 Minaninakano-cho, Nishishich(iyo, Shimogyo-ku, Kyoto 600, Japan bDepartment of Radiology, Kyoto Minami Hospital, 8 Minaminakano.cho, Nishishichijyo, Shimogyo-ku, Kyoto 600, Japan ¢Central Research Institute, Shimadzu Corporation, 1 Nishinokyo-Kuwabaracho, Nakagyo-ku, Kyoto 604, Japan
Received 28 February 1995; revision received 1 May 1995; accepted 15 May 1994
Abstract
To assess the accuracy of automatic extraction of the left ventricular inner contour on contrast-enhanced ultrafast cine magnetic resonance (MR) images, we compared the values obtained by this method with those obtained using intravenous digital subtraction left ventriculography. High-quality single breath-hold contrast-enhanced ultrafast cine MR images were obtained in all cardiac phases on horizontal and vertical long axis sections of the left ventricle. For ultrafast cine MR imaging, a phase-rewind gradientecho (rewind-SMASH) sequence was used. Automatic extraction of the left ventricular inner contour on contrast-enhanced ultrafast cine MR images was performed in all cardiac phases. High-quality left ventricular images of the horizontal long axis section were obtained in 127 of 160 patients (79%). The automatic extraction of the left ventricular contour was easily performed on high-quality images with very short processing time (4 s/frame). The values for left ventricular volumes obtained with the automatic extraction method on contrast-enhanced ultrafast cine MR imaging were correlated well with those obtained with the manual extraction method and IV-DSA in high quality cardiac images. The biplane modified Simpson's method using automatic extraction is an accurate and highly reproducible method for evaluating left ventricular volumes. Keywords: Magnetic resonance (MR) imaging, heart; Magnetic resonance (MR), technique; Heart, MRI
1. Introduction
In the evaluation o f cardiac function by magnetic resonance (MR) imaging, standard gradient-echo cine MR imaging is a lengthy procedure, and the quality of the images is sometimes impaired by ghosting and blurring due to motion artifacts. The MR imaging technology of data acquisition has recently advanced, and imaging time has been reduced to <50 ms. Currently, the ultrafast MR imaging methods include snapshot MR imaging [1-5], echo planar imaging [6-8] and segmented turbo-FLASH [9,10]. The left ventricular volume curve by snapshot MR imaging is inaccurate because the data acquisition time is over 300 ms/frame. In echo planar imaging, the machine system of MR im*Corresponding author, Tel.: +81 75 3127361; Fax: +81 75 3146290.
aging should be generally improved for a large gradient echo field. The segmented k-space method, in which the data acquisition is divided into 16 segments, is realized by a 48 ms-acquisition window with a total imaging interval of 16 heart beats [10]. In this method, the left ventricular volume and volume curve can be accurately calculated. With single breath-hold ultrafast cine MR imaging, although section images of the left ventricle cannot be obtained without the use of contrast material, high quality MR images can be obtained during firstpass circulation of the contrast agent Ill]. While contrast-enhanced breath-hold ultrafast cine MR imaging has been shown to provide accurate cardiac images with high reproducibility [1 I], the data analysis method still depends on the operator's manual extraction of the inner left ventricular contour, and subjectivity therefore cannot be excluded. In the present study, we performed automatic extraction of the left ventricular inner con-
07204)48X/95/$09.50 © 1995 Elsevier ScienceIreland Ltd. All rights reserved SSDI 0720-048X(95)00641-3
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
127
tour based on contrast-enhanced ultrafast cine MR images, and evaluated the accuracy of measurement of left ventricular volume by the modified biplane Simpson's method, according to automatic extraction by comparison with biplane intravenous digital subtraction left ventriculography (IV-DSA).
i -i
(i.*',
:__
image
2. Material and methods
The single breath-hold contrast-enhanced ultrafast cine MR imaging was performed in 160 patients on the horizontal long axis and vertical long axis sections of the left ventricle. Ninety patients had ischemic heart disease, 35 had hypertensive heart disease, 20 had valvular heart disease, seven had dilated cardiomyopathy and eight had other heart diseases. Eighty-two patients also underwent IV-DSA within 2 months after MR examination. During this study period, no patient presented any new cardiac event. All subjects signed an informed consent and the study was approved by the Human Ethical Provision in the Research Committee of Kyoto Minami Hospital.
2.1. Single breath-hold contrast-enhanced ultrafast cine MR imaging Images were obtained with a 1.5-T whole-body MR imaging system (Shimadzu SMT-150X; Shimadzu Corporation, Kyoto, Japan). The data were acquired by using a rewind-SMASH (Short Minimum Angle SHot) sequence [11] with TR of 8 ms, TE of 3.2 ms and flip angle of 25°. The acquisition was divided into 16 segments, and one segment consisted of six phaseencoding steps. A 128 x 96 matrix requiring 16 heart beats was needed. The time interval for one frame was 48 ms, the section thickness was 10 mm, the number of excitations was one, and the size of the rectangular field of view was 200 x 200 mm. After confirmation of accurate R-wave triggering on the left ventricular horizontal and vertical long axis sections during data acquisition on single breath-hold ultrafast cine MR imaging, Gd-DTPA (0.05 mmol/kg) was injected i.v., and single breath-hold ultrafast cine MR images in the two sections were again obtained during first-pass cardiac circulation of the contrast agent. 2.2. Evaluation of MR image quality The quality of cardiac images obtained by single breath-hold contrast-enhanced ultrafast cine MR imaging was evaluated as follows by three observers (two radiologists and one cardiologist): Good: the left ventricular inner contour is easily extracted by both automatic and manual methods. Fair: the left ventricular inner contour is difficult to extract in all cardiac images, and can only be extracted in end-diastolic and end-systolic images. Poor: the left ventricular inner contour cannot be extracted.
Rsn
Lsn
Rs~
Fig. 1. Scheme of automatic extraction of the left ventricular inner contour. The operator must first input three contour points (PINI, PIN2, PIN3) on the first image (usually the end-diastolic image), which indicate the two ends of the mitral ring and the apex. A radial scan line from the center point calculated using the three points is then created, and by scanning each scan line from the center point to the end of the line, all of the candidate points which may be contour points are identified.
2.3. Automatic extraction of left ventricular inner contours The operator must initially input the three contour points on the first image (usually the end-diastolic image), which indicate the bilateral ends of the mitral ring and left ventricular apex. The mitral ring contour is obtained by connecting both end points with a straight line. The radial scan lines from the center point are then calculated using the three points. A candidate point, which may be a contour point, is defined as a point at which the image density slope value along the scan line is at local minimum, or a point at which the image density value is lower than the automatically calculated threshold value. Finally, the combination of candidate points with the best fit against the scan lines (except the mitral ring and apex) are selected by evaluating the smoothness of each set of points. On the second image, the scan lines created in the first image processing are adopted again, and the candidate points are identified using the same method as in the first image processing. The same process is repeated successively for other images, completing the automatic extraction process (Fig. 1). In this study, the automatic extraction of the left ventricular inner contour was applied to the horizontal long
128
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
a
b
c
d
Fig. 2. The sequencesshow automatic extraction of left ventricularend-diastolicand end-systolicimages on the horizontallong axis section(a,c) and vertical long axis section (b,d).
axis sections and the vertical long axis sections in all cardiac phases obtained by contrast-enhanced single breath-hold ultrafast cine MR imaging. The processing time of automatic extraction was 4 s on each cardiac frame. The automatic extraction was performed only once, and if automatic analysis was mistaken, manual extraction was applied to each cardiac frame. 2.4. Manual extraction of left ventricular inner contours The inner edge of the left ventricular cavity was detected by manual tracing with a track-ball, and the length of the long axis of the cavity was calculated from the distance between the center point of the mitral ring
and inner apex. The papillary muscles were excluded on a manual tracing of the inner contour of the left ventricle. The manual extraction was performed on the horizontal long axis section of the left ventricle. Manual tracings were performed by three observers (two radiologists and one cardiologist). 2.5. Analysis of left ventricular hemodynamics The end-diastolic volume was constructed from the first frame on the horizontal and vertical long axis sections of the left ventricle (Fig. 2a,b), and the end-systolic volume was constructed from each frame at the T-wave terminal on the same cardiac sections (Fig. 2c,d). The
129
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132 150
ml/m z
EDV I modified Simpson's method
biplane
100
ESV I biplane modified Simpson' s method
ml/m z
==
v
i
e
n = 60 y = 0.875x r - O. 9 3 0
+ 4. 11
p < 0.001
mile z i
0
0
150
LV-sraphy
S •
•
n
= 60
y = 0.920x - 0.88
o•
r = 0.969
p < 0.001 =l/m z I 100
LV-gr~v
Fig. 3. Graphs showing values for the left ventricular end-diastolic volume index (left panel) and end-systolic volume index (right panel) obtained using contrast-enhanced ultrafast cine MR imaging and a biplane modified Simpson's method plotted against those obtained using IV-DSA of the left ventricle and the biplane area-length method. EDVI, end-diastolic volume index; ESVI, end-systolic volume index; UF, ultrafast; LV-graphy, left ventricular digital subtraction angiography.
contours of the left ventricular cavity were automatically detected, and left ventricular volume was measured using the biplane modified Simpson's method [12]. The left ventricular volume curve was automatically constructed by the monoplane area-length method on all cardiac frames [13]. On the basis of the left ventricular volume curve, peak ejection rate, peak filling rate and other functional parameters were measured. In this study, the end-diastolic volume index and end-systolic volume index obtained by the biplane modified Simpson's method were compared with those obtained with the biplane IV-DSA of the left ventricle. The peak ejec-
Peak
Eject
ion
tion rate and peak filling rate measured on the automatically produced left ventricular volume curve were compared with the left ventricular volume curve manually produced by the monoplane area-length method [13]. Peak ejection and peak filling rates, obtained with the manual method by three observers, were averaged.
2.6. IV digital subtraction left ventriculography A high-flow catheter was inserted into a central vein, and contrast medium (iopamidol-370; iodine content 370 mg/ml) was injected rapidly (0.4 ml/kg/s). Data were
Peak
Rate
1/~
F i l l i n g
Rate
l/sec
A
o
z
i
.o •
?
o• oe
• ooo
oo_
"°~'_
• S=2:.,...
• : ~ , r ,a: •
• •
•
J°ee /
•
o•
•
•
811
" ~
•.
i,
"
i
.
..
•
•:
".~.'o .; L '* ' - -
"
••
•
o• •
o•
e • • ~o o o
• • o• • • ::," ."./f..
• • °•
n = 127
..,,dr8-~ -],,- •
y = O.904x .F 0.64
f t
r = 0.816 p < 0.001
*'•
• 1/sec
. •
"
•
__ n = 127 y : 0.SS3x
• •
p < 0.001
l/sec
I
UF-c ine-I~l (manual)
7
~ O. 865
r = 0.697
! UF-c i ne-itl
(mmtml)
7
Fig. 4. Graphs showing values for the left ventricular peak ejection rate (left panel) and the peak filling rate (right panel) obtained using an automatic extraction method plotted against those obtained using a manual extraction method of the left ventricular inner contour. UF, ultrafast.
130
K. Matsumuraet al./ European Journalof Radiology 20 (1995) 126-132
acquired in the right anterior oblique 30° and left anterior oblique 60° planes with DSA-equipment (Toshiba DFP-50A, Toshiba Co, Tokyo, Japan). The image matrix was 256 x 256, at 8 bit, and 30 frames/s [14].
3. Results 3.1. Image quality of single breath-hold contrast-enhanced ultrafast cine MR imaging The quality of cardiac images was evaluated by three observers. Of the 160 patients in whom horizontal long axis sections of the left ventricle were acquired, good images were obtained in 127 (79%), fair images in 20 (13%), and poor images in 13 (8%). On the vertical long axis section of the left ventricle, good images were obtained in 119 (74%), fair images in 22 (14%), and poor images in 19 (12%). The left ventricular end-diastolic and endsystolic volumes could be obtained by the biplane modified Simpson's method in 141 patients (88%) with automatic extraction of the left ventricular inner contour. 3.2. Left ventricular volumes obtained by the automatic extraction method and the modified biplane Simpson's rule method Comparison of left ventricular end-diastolic and endsystolic volumes with ultrafast cine MR imaging and IVDSA was evaluated in 60 patients in whom good quality images were obtained. As shown in Fig. 3, the left ventricular end-diastolic and end-systolic volumes obtained using the modified biplane Simpson's method in ultrafast cine MR imaging and biplane area-length method in IV-DSA showed a high correlation (enddiastolic: y = 0.875x + 4.11, r = 0.930, P < 0.001; endsystolic, y - 0.920x - 0.88, r = 0.969, P < 0.001). The values for the left ventricular ejection fraction obtained by the two methods also showed a good correlation (y = 0.892x + 5.02, r = 0.882, P < 0.001). 3.3. Left ventricular volume curve obtained by the automatic extraction method and the manual extraction method On the good left ventricular images in all cardiac phases, the left ventricular volume curves were similar for the automatic extraction method and the manual extraction method. The left ventricular volume curve was obtained in 127 patients by automatic and manual extraction of the left ventricular inner contour. As shown in Fig. 4, the left ventricular peak ejection rate and peak filling rate obtained by the left ventricular volume curve in the automatic extraction method and the manual extraction method showed a statistically significant correlation (peak ejection rate: y = 0.904x + 0.64, r = 0.816, P < 0.001; peak filling rate: y = 0.853x + 0.865, r = 0.697, P < 0.001).
4. Discussion The more slowly acquired ECG-gated conventional cine MR images have ghosting and blurring caused by motion artifacts, and the left ventricular images are unsatisfactory. In conventional cine MR imaging, the data acquisition and image analysis time (manual method) are very long (acquisition time: 8.5 min/section within a 1000-ms R-R interval, 256 x 256 matrix and two excitations; analysis time: 90-180 s/frame by a trained observer), so that clinical use of this method has remained limited. However, the clinical usefulness of enddiastolic and end-systolic volume of the left ventricle obtained by conventional cine MR imaging has already been established [15-18], the reproducibility of this method is high [19,20], and temporal resolution is satisfactory. The reliability of the monoplane area-length method is decreased when the left ventricle is markedly deformed [21], in which case the biplane area-length method or the Simpson's method are performed to increase accuracy [17,19,20]. Also in ECG-gated conventional cine MR imaging, data acquisition and analysis times are long, and neither the biplane area-length method nor the Simpson's method have become widely used in clinical practice. With single breath-hold ultrafast cine MR imaging (rewind-SMASH), high quality MR images can be obtained during first-pass circulation of the contrast material [11]. The segmented turbo-FLASH has enabled acquisition of highresolution cine images during one breath-hold [9,10]. However, with this method, a severe striping artifact is noted at an increased flip angle (> 30°). To eliminate the striping artifact, we developed a new technique that uses the rewind-SMASH sequence. With this method, the striping artifact does not appear, even at larger flip angles. In addition, use of dummy RF pulses suppresses the flashing phenomenon observed on images obtained during the first cardiac phase and produces uniformity of image contrast in each cardiac phase [11]. The acquisition window per segment was 48 ms with the rewindSMASH sequence, but this interval was long and unsatisfactory for obtaining the left ventricular volume curve. In echo planar imaging, high temporal resolution was achieved, but required special hardware [8]. Recently, the automated detection of left ventricular epi- and endocardial contours has been reported in the short axis section [22], but not in the long axis section on ultrafast cine MR imaging. We recently developed the automatic method of extraction of the left ventricular inner contour on cine MR images, which requires the manual input of only three points on the first image and requires no manual operation for other images. In this automatic process, the left ventricular volumes are calculated and the volume curve is easily constructed. The automatic extraction of the left ventricular inner contour requires only 4 s per cardiac frame and total
K. Matsumura et al. /European Journal of Radiology 20 (1995) 126-132
time for acquisition and analysis was 96 s/section at 60 beats/min heart rate. The success rate of automatic extraction of the end-diastolic and end-systolic phase was very high. On high quality cardiac images, left ventricular volume obtained by automatic extraction and the modified biplane Simpson's method showed a high correlation with that obtained using IV-DSA and the biplane area-length method. But the former slightly underestimated the left ventricular volume compared to the latter. For this reason, the left ventriculography in DSA is a projected configuration and includes the ventricular papillary muscles, while that in MR images excludes the papillary muscles [23]. The left ventricular volume is underestimated in the automatic extraction method of MR imaging for high threshold level of differential density profile curve. The method of automatic extraction is based on image density, so that definition of the left ventricular contour by automatic extraction is distracted under signal void and when marked deformity of the left ventricle is present. In such cases, the left ventricular contour is manually retraced in the automatically mistraced area, but these cases are infrequent. The peak ejection rate and peak filling rate as a left ventricular performance obtained using the automatic extraction and the monoplane area-length method showed a statistically significant correlation with that obtained using the manual extraction method, but the correlation coefficient of peak filling rate was slightly decreased.
4.1. Study limitations There are two major limitations to our method. First, the time interval between cardiac frames was relatively long (48 ms) as compared to conventional cine MR imaging (30 ms). This interval is not sufficient for recognition of an accurate end-systolic point. The other methods are required for high temporal resolution (e.g. breath-hold EPI). Second, intravenous contrast agent is necessary for imaging in our method. The left ventricular images are obtained without contrast agent on our ultrafast MR imaging in the short axis sections, and automatic extraction of the left ventricular inner contours is possible. The cardiac chamber volumes can be calculated by the true Simpson's method in these short axis sections. However, accuracy of this method is not sufficient because the slice thickness is > 10 mm. 5. Conclusions
The automatic extraction method of the left ventricular inner contour on contrast-enhanced ultrafast cine MR images is accurate and reproducible. Left ventricular volumes and the ejection fraction are rapidly and accurately evaluated by the biplane modified Simpson's method. The volume curve of the left ventricle is easily and quickly obtained using the monoplane area-length
131
method based on automatic extraction of the left ventricular inner contour. References [1] Hasse A. Snapshot FLASH MRI: application to TI, T2, and chemical-shift imaging. Magn Reson Med 1990; 13: 77-89. [2l Atkins DJ, Burstein D, Edelman RR. First-pass cardiac pcrfusion: evaluation with ultrafast MR imaging. Radiology 1990; 174: 757-762. [3] Fram J, Merboldt KD, Bruhn H, Gyngell ML, Hanicke W, Chien D. 0.3 second FLASH MRI of the human heart. Magn Reson Med 1990; 13: 150-157. [4] Matthaei D, Hasse A, Henrich D, Duhmke E. Cardiac and vascular imaging with an MR snapshot technique. Radiology 1990; 177: 527-532. [5] Holsinger AE, Wright RC, Riederer SJ, Farzaneh F, Grimm RC, Maier JK. Real-time interactive magnetic resonance imaging. Magn Reson Med 1990; 14: 547-553. [6] Mansfield P. Multiplanar image formation using NMR spinechoes. J Phys C Solid State Phys 1977; 10: L55-L58. [7] Rosen BR, Bclliveau JW, Vevea JM, Brady TJ. Perfusion imaging with NMR contrast agents. Magn Reson Med 1990; 14: 249-265. [8] Worthington BS, Mansfield P. The clinical application of echo planar imaging in neuroradiology. Neuroradiology 1990; 32: 367-370. [9] Edelman RR, Wailner B, Singer A, Atkinson D J, Saini S. Segmented turboFLASH: method for breath-hold MR imaging of the liver with flexible contrast. Radiology 1990; 177: 515-512. ll0] Atkinson DJ, Edelman RR. Cineangiography of the heart in a single breath hold with a segmented turboFLASH sequence. Radiology 1991; 178: 357-360. [11] Matsumura K, Nakase E, Haiyama T, Takeo K, Shimizu K, Yamasaki K, Kohno K. Determination of cardiac ejection fraction and left ventricular volume: contrast-enhanced ultrafast cine-MR imaging vs. IV digital subtraction ventriculography. AJR 1993; 160: 979-985. [12] Chapman CB, Baker O, Reynolds J. Use of biplane cinefluorography for measurement of ventricular volume. Circulation 1958; 18: ! 105-1 ! 17. [13] Dodge HT, Sandier H, Baxley WA. Usefulness and limitations of radiographic methods for determining left ventricular volume. Am J Cardiol 1966; 18: 10-24. [141 Matsumura K, Nakase E, Kawai I, Saito T. Clinical evaluation of cardiac intravenous digital subtraction angiographyfunctional analysis of the four cardiac chambers. Med Rev 1992; 41: 1-13. [15] Sechtem U, Pflugfelder PW, White RD. Cine MR imaging: potential for the evaluation of cardiovascular function. A JR 1987; 148: 239-246. [16] Higgins CB, Holt W, Pflugfelder P, Sechtem U. Functional evaluation of the heart with magnetic resonance imaging. Magn Resort Med 1988; 6: 121-139. [17] Buser PT, Auffermann W, Holt WW. Noninvasive evaluation of global left ventricular function with the use of cine nuclear magnetic resonance. J Am Coil Cardiol 1989; 13: 1294-1300. [18] Cranney GB, Lotan CS, Dean L, Baxley W, Bouchard A, Pohost GM. Left ventricular volume measurement using cardiac axis NMR imaging-validation by calibrated ventricular angiography. Circulation 1990; 82: 154-163. [19] Semelka RC, Tomei E, Wagner S. Normal left ventricular dimensions and functions: interstudy reproducibility of measurements with cine MR. Radiology 1990; 174: 763-768.
132
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
[20] Benjelloun H, Cranney GB, Kirk KA. Interstudy reproducibility of biplane cine nuclear magnetic resonance measurements of left ventricular function. Am J Cardiol 1991; 67: 1413-1420. [211 Teichholz LE, Kreulen T, Herman MW. Problems in echocardiographic volume determinations: echocardiographicangiographic correlations in the presence or absence of asynergy. Am J Cardiol 1976; 37:7-1 I.
[22] van der Gecst R J, Jansen E, Builer VGM, Reiber JHC. Automated detection of left ventricular epi- and endocardial contours in short-axis MR images. Comp Cardiol 1994; 33-36. [23] Rackely CE. Quantitative evaluation of left ventricular function by radiographic techniques. Circulation 1976; 54: 862-879.
ELSEVIER
European Journal of Radiology 20 (1995) 126-132
Automatic left ventricular volume measurements on contrast-enhanced ultrafast cine magnetic resonance imaging Kentaro Matsumura *a, Emiko Nakase a, Tohru Haiyama b, Shinichi Utsunomiya c aDepartment of Internal Medicine, Kyoto Minami Hospital, 8 Minaninakano-cho, Nishishich(iyo, Shimogyo-ku, Kyoto 600, Japan bDepartment of Radiology, Kyoto Minami Hospital, 8 Minaminakano.cho, Nishishichijyo, Shimogyo-ku, Kyoto 600, Japan ¢Central Research Institute, Shimadzu Corporation, 1 Nishinokyo-Kuwabaracho, Nakagyo-ku, Kyoto 604, Japan
Received 28 February 1995; revision received 1 May 1995; accepted 15 May 1994
Abstract
To assess the accuracy of automatic extraction of the left ventricular inner contour on contrast-enhanced ultrafast cine magnetic resonance (MR) images, we compared the values obtained by this method with those obtained using intravenous digital subtraction left ventriculography. High-quality single breath-hold contrast-enhanced ultrafast cine MR images were obtained in all cardiac phases on horizontal and vertical long axis sections of the left ventricle. For ultrafast cine MR imaging, a phase-rewind gradientecho (rewind-SMASH) sequence was used. Automatic extraction of the left ventricular inner contour on contrast-enhanced ultrafast cine MR images was performed in all cardiac phases. High-quality left ventricular images of the horizontal long axis section were obtained in 127 of 160 patients (79%). The automatic extraction of the left ventricular contour was easily performed on high-quality images with very short processing time (4 s/frame). The values for left ventricular volumes obtained with the automatic extraction method on contrast-enhanced ultrafast cine MR imaging were correlated well with those obtained with the manual extraction method and IV-DSA in high quality cardiac images. The biplane modified Simpson's method using automatic extraction is an accurate and highly reproducible method for evaluating left ventricular volumes. Keywords: Magnetic resonance (MR) imaging, heart; Magnetic resonance (MR), technique; Heart, MRI
1. Introduction
In the evaluation o f cardiac function by magnetic resonance (MR) imaging, standard gradient-echo cine MR imaging is a lengthy procedure, and the quality of the images is sometimes impaired by ghosting and blurring due to motion artifacts. The MR imaging technology of data acquisition has recently advanced, and imaging time has been reduced to <50 ms. Currently, the ultrafast MR imaging methods include snapshot MR imaging [1-5], echo planar imaging [6-8] and segmented turbo-FLASH [9,10]. The left ventricular volume curve by snapshot MR imaging is inaccurate because the data acquisition time is over 300 ms/frame. In echo planar imaging, the machine system of MR im*Corresponding author, Tel.: +81 75 3127361; Fax: +81 75 3146290.
aging should be generally improved for a large gradient echo field. The segmented k-space method, in which the data acquisition is divided into 16 segments, is realized by a 48 ms-acquisition window with a total imaging interval of 16 heart beats [10]. In this method, the left ventricular volume and volume curve can be accurately calculated. With single breath-hold ultrafast cine MR imaging, although section images of the left ventricle cannot be obtained without the use of contrast material, high quality MR images can be obtained during firstpass circulation of the contrast agent Ill]. While contrast-enhanced breath-hold ultrafast cine MR imaging has been shown to provide accurate cardiac images with high reproducibility [1 I], the data analysis method still depends on the operator's manual extraction of the inner left ventricular contour, and subjectivity therefore cannot be excluded. In the present study, we performed automatic extraction of the left ventricular inner con-
07204)48X/95/$09.50 © 1995 Elsevier ScienceIreland Ltd. All rights reserved SSDI 0720-048X(95)00641-3
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
127
tour based on contrast-enhanced ultrafast cine MR images, and evaluated the accuracy of measurement of left ventricular volume by the modified biplane Simpson's method, according to automatic extraction by comparison with biplane intravenous digital subtraction left ventriculography (IV-DSA).
i -i
(i.*',
:__
image
2. Material and methods
The single breath-hold contrast-enhanced ultrafast cine MR imaging was performed in 160 patients on the horizontal long axis and vertical long axis sections of the left ventricle. Ninety patients had ischemic heart disease, 35 had hypertensive heart disease, 20 had valvular heart disease, seven had dilated cardiomyopathy and eight had other heart diseases. Eighty-two patients also underwent IV-DSA within 2 months after MR examination. During this study period, no patient presented any new cardiac event. All subjects signed an informed consent and the study was approved by the Human Ethical Provision in the Research Committee of Kyoto Minami Hospital.
2.1. Single breath-hold contrast-enhanced ultrafast cine MR imaging Images were obtained with a 1.5-T whole-body MR imaging system (Shimadzu SMT-150X; Shimadzu Corporation, Kyoto, Japan). The data were acquired by using a rewind-SMASH (Short Minimum Angle SHot) sequence [11] with TR of 8 ms, TE of 3.2 ms and flip angle of 25°. The acquisition was divided into 16 segments, and one segment consisted of six phaseencoding steps. A 128 x 96 matrix requiring 16 heart beats was needed. The time interval for one frame was 48 ms, the section thickness was 10 mm, the number of excitations was one, and the size of the rectangular field of view was 200 x 200 mm. After confirmation of accurate R-wave triggering on the left ventricular horizontal and vertical long axis sections during data acquisition on single breath-hold ultrafast cine MR imaging, Gd-DTPA (0.05 mmol/kg) was injected i.v., and single breath-hold ultrafast cine MR images in the two sections were again obtained during first-pass cardiac circulation of the contrast agent. 2.2. Evaluation of MR image quality The quality of cardiac images obtained by single breath-hold contrast-enhanced ultrafast cine MR imaging was evaluated as follows by three observers (two radiologists and one cardiologist): Good: the left ventricular inner contour is easily extracted by both automatic and manual methods. Fair: the left ventricular inner contour is difficult to extract in all cardiac images, and can only be extracted in end-diastolic and end-systolic images. Poor: the left ventricular inner contour cannot be extracted.
Rsn
Lsn
Rs~
Fig. 1. Scheme of automatic extraction of the left ventricular inner contour. The operator must first input three contour points (PINI, PIN2, PIN3) on the first image (usually the end-diastolic image), which indicate the two ends of the mitral ring and the apex. A radial scan line from the center point calculated using the three points is then created, and by scanning each scan line from the center point to the end of the line, all of the candidate points which may be contour points are identified.
2.3. Automatic extraction of left ventricular inner contours The operator must initially input the three contour points on the first image (usually the end-diastolic image), which indicate the bilateral ends of the mitral ring and left ventricular apex. The mitral ring contour is obtained by connecting both end points with a straight line. The radial scan lines from the center point are then calculated using the three points. A candidate point, which may be a contour point, is defined as a point at which the image density slope value along the scan line is at local minimum, or a point at which the image density value is lower than the automatically calculated threshold value. Finally, the combination of candidate points with the best fit against the scan lines (except the mitral ring and apex) are selected by evaluating the smoothness of each set of points. On the second image, the scan lines created in the first image processing are adopted again, and the candidate points are identified using the same method as in the first image processing. The same process is repeated successively for other images, completing the automatic extraction process (Fig. 1). In this study, the automatic extraction of the left ventricular inner contour was applied to the horizontal long
128
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
a
b
c
d
Fig. 2. The sequencesshow automatic extraction of left ventricularend-diastolicand end-systolicimages on the horizontallong axis section(a,c) and vertical long axis section (b,d).
axis sections and the vertical long axis sections in all cardiac phases obtained by contrast-enhanced single breath-hold ultrafast cine MR imaging. The processing time of automatic extraction was 4 s on each cardiac frame. The automatic extraction was performed only once, and if automatic analysis was mistaken, manual extraction was applied to each cardiac frame. 2.4. Manual extraction of left ventricular inner contours The inner edge of the left ventricular cavity was detected by manual tracing with a track-ball, and the length of the long axis of the cavity was calculated from the distance between the center point of the mitral ring
and inner apex. The papillary muscles were excluded on a manual tracing of the inner contour of the left ventricle. The manual extraction was performed on the horizontal long axis section of the left ventricle. Manual tracings were performed by three observers (two radiologists and one cardiologist). 2.5. Analysis of left ventricular hemodynamics The end-diastolic volume was constructed from the first frame on the horizontal and vertical long axis sections of the left ventricle (Fig. 2a,b), and the end-systolic volume was constructed from each frame at the T-wave terminal on the same cardiac sections (Fig. 2c,d). The
129
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132 150
ml/m z
EDV I modified Simpson's method
biplane
100
ESV I biplane modified Simpson' s method
ml/m z
==
v
i
e
n = 60 y = 0.875x r - O. 9 3 0
+ 4. 11
p < 0.001
mile z i
0
0
150
LV-sraphy
S •
•
n
= 60
y = 0.920x - 0.88
o•
r = 0.969
p < 0.001 =l/m z I 100
LV-gr~v
Fig. 3. Graphs showing values for the left ventricular end-diastolic volume index (left panel) and end-systolic volume index (right panel) obtained using contrast-enhanced ultrafast cine MR imaging and a biplane modified Simpson's method plotted against those obtained using IV-DSA of the left ventricle and the biplane area-length method. EDVI, end-diastolic volume index; ESVI, end-systolic volume index; UF, ultrafast; LV-graphy, left ventricular digital subtraction angiography.
contours of the left ventricular cavity were automatically detected, and left ventricular volume was measured using the biplane modified Simpson's method [12]. The left ventricular volume curve was automatically constructed by the monoplane area-length method on all cardiac frames [13]. On the basis of the left ventricular volume curve, peak ejection rate, peak filling rate and other functional parameters were measured. In this study, the end-diastolic volume index and end-systolic volume index obtained by the biplane modified Simpson's method were compared with those obtained with the biplane IV-DSA of the left ventricle. The peak ejec-
Peak
Eject
ion
tion rate and peak filling rate measured on the automatically produced left ventricular volume curve were compared with the left ventricular volume curve manually produced by the monoplane area-length method [13]. Peak ejection and peak filling rates, obtained with the manual method by three observers, were averaged.
2.6. IV digital subtraction left ventriculography A high-flow catheter was inserted into a central vein, and contrast medium (iopamidol-370; iodine content 370 mg/ml) was injected rapidly (0.4 ml/kg/s). Data were
Peak
Rate
1/~
F i l l i n g
Rate
l/sec
A
o
z
i
.o •
?
o• oe
• ooo
oo_
"°~'_
• S=2:.,...
• : ~ , r ,a: •
• •
•
J°ee /
•
o•
•
•
811
" ~
•.
i,
"
i
.
..
•
•:
".~.'o .; L '* ' - -
"
••
•
o• •
o•
e • • ~o o o
• • o• • • ::," ."./f..
• • °•
n = 127
..,,dr8-~ -],,- •
y = O.904x .F 0.64
f t
r = 0.816 p < 0.001
*'•
• 1/sec
. •
"
•
__ n = 127 y : 0.SS3x
• •
p < 0.001
l/sec
I
UF-c ine-I~l (manual)
7
~ O. 865
r = 0.697
! UF-c i ne-itl
(mmtml)
7
Fig. 4. Graphs showing values for the left ventricular peak ejection rate (left panel) and the peak filling rate (right panel) obtained using an automatic extraction method plotted against those obtained using a manual extraction method of the left ventricular inner contour. UF, ultrafast.
130
K. Matsumuraet al./ European Journalof Radiology 20 (1995) 126-132
acquired in the right anterior oblique 30° and left anterior oblique 60° planes with DSA-equipment (Toshiba DFP-50A, Toshiba Co, Tokyo, Japan). The image matrix was 256 x 256, at 8 bit, and 30 frames/s [14].
3. Results 3.1. Image quality of single breath-hold contrast-enhanced ultrafast cine MR imaging The quality of cardiac images was evaluated by three observers. Of the 160 patients in whom horizontal long axis sections of the left ventricle were acquired, good images were obtained in 127 (79%), fair images in 20 (13%), and poor images in 13 (8%). On the vertical long axis section of the left ventricle, good images were obtained in 119 (74%), fair images in 22 (14%), and poor images in 19 (12%). The left ventricular end-diastolic and endsystolic volumes could be obtained by the biplane modified Simpson's method in 141 patients (88%) with automatic extraction of the left ventricular inner contour. 3.2. Left ventricular volumes obtained by the automatic extraction method and the modified biplane Simpson's rule method Comparison of left ventricular end-diastolic and endsystolic volumes with ultrafast cine MR imaging and IVDSA was evaluated in 60 patients in whom good quality images were obtained. As shown in Fig. 3, the left ventricular end-diastolic and end-systolic volumes obtained using the modified biplane Simpson's method in ultrafast cine MR imaging and biplane area-length method in IV-DSA showed a high correlation (enddiastolic: y = 0.875x + 4.11, r = 0.930, P < 0.001; endsystolic, y - 0.920x - 0.88, r = 0.969, P < 0.001). The values for the left ventricular ejection fraction obtained by the two methods also showed a good correlation (y = 0.892x + 5.02, r = 0.882, P < 0.001). 3.3. Left ventricular volume curve obtained by the automatic extraction method and the manual extraction method On the good left ventricular images in all cardiac phases, the left ventricular volume curves were similar for the automatic extraction method and the manual extraction method. The left ventricular volume curve was obtained in 127 patients by automatic and manual extraction of the left ventricular inner contour. As shown in Fig. 4, the left ventricular peak ejection rate and peak filling rate obtained by the left ventricular volume curve in the automatic extraction method and the manual extraction method showed a statistically significant correlation (peak ejection rate: y = 0.904x + 0.64, r = 0.816, P < 0.001; peak filling rate: y = 0.853x + 0.865, r = 0.697, P < 0.001).
4. Discussion The more slowly acquired ECG-gated conventional cine MR images have ghosting and blurring caused by motion artifacts, and the left ventricular images are unsatisfactory. In conventional cine MR imaging, the data acquisition and image analysis time (manual method) are very long (acquisition time: 8.5 min/section within a 1000-ms R-R interval, 256 x 256 matrix and two excitations; analysis time: 90-180 s/frame by a trained observer), so that clinical use of this method has remained limited. However, the clinical usefulness of enddiastolic and end-systolic volume of the left ventricle obtained by conventional cine MR imaging has already been established [15-18], the reproducibility of this method is high [19,20], and temporal resolution is satisfactory. The reliability of the monoplane area-length method is decreased when the left ventricle is markedly deformed [21], in which case the biplane area-length method or the Simpson's method are performed to increase accuracy [17,19,20]. Also in ECG-gated conventional cine MR imaging, data acquisition and analysis times are long, and neither the biplane area-length method nor the Simpson's method have become widely used in clinical practice. With single breath-hold ultrafast cine MR imaging (rewind-SMASH), high quality MR images can be obtained during first-pass circulation of the contrast material [11]. The segmented turbo-FLASH has enabled acquisition of highresolution cine images during one breath-hold [9,10]. However, with this method, a severe striping artifact is noted at an increased flip angle (> 30°). To eliminate the striping artifact, we developed a new technique that uses the rewind-SMASH sequence. With this method, the striping artifact does not appear, even at larger flip angles. In addition, use of dummy RF pulses suppresses the flashing phenomenon observed on images obtained during the first cardiac phase and produces uniformity of image contrast in each cardiac phase [11]. The acquisition window per segment was 48 ms with the rewindSMASH sequence, but this interval was long and unsatisfactory for obtaining the left ventricular volume curve. In echo planar imaging, high temporal resolution was achieved, but required special hardware [8]. Recently, the automated detection of left ventricular epi- and endocardial contours has been reported in the short axis section [22], but not in the long axis section on ultrafast cine MR imaging. We recently developed the automatic method of extraction of the left ventricular inner contour on cine MR images, which requires the manual input of only three points on the first image and requires no manual operation for other images. In this automatic process, the left ventricular volumes are calculated and the volume curve is easily constructed. The automatic extraction of the left ventricular inner contour requires only 4 s per cardiac frame and total
K. Matsumura et al. /European Journal of Radiology 20 (1995) 126-132
time for acquisition and analysis was 96 s/section at 60 beats/min heart rate. The success rate of automatic extraction of the end-diastolic and end-systolic phase was very high. On high quality cardiac images, left ventricular volume obtained by automatic extraction and the modified biplane Simpson's method showed a high correlation with that obtained using IV-DSA and the biplane area-length method. But the former slightly underestimated the left ventricular volume compared to the latter. For this reason, the left ventriculography in DSA is a projected configuration and includes the ventricular papillary muscles, while that in MR images excludes the papillary muscles [23]. The left ventricular volume is underestimated in the automatic extraction method of MR imaging for high threshold level of differential density profile curve. The method of automatic extraction is based on image density, so that definition of the left ventricular contour by automatic extraction is distracted under signal void and when marked deformity of the left ventricle is present. In such cases, the left ventricular contour is manually retraced in the automatically mistraced area, but these cases are infrequent. The peak ejection rate and peak filling rate as a left ventricular performance obtained using the automatic extraction and the monoplane area-length method showed a statistically significant correlation with that obtained using the manual extraction method, but the correlation coefficient of peak filling rate was slightly decreased.
4.1. Study limitations There are two major limitations to our method. First, the time interval between cardiac frames was relatively long (48 ms) as compared to conventional cine MR imaging (30 ms). This interval is not sufficient for recognition of an accurate end-systolic point. The other methods are required for high temporal resolution (e.g. breath-hold EPI). Second, intravenous contrast agent is necessary for imaging in our method. The left ventricular images are obtained without contrast agent on our ultrafast MR imaging in the short axis sections, and automatic extraction of the left ventricular inner contours is possible. The cardiac chamber volumes can be calculated by the true Simpson's method in these short axis sections. However, accuracy of this method is not sufficient because the slice thickness is > 10 mm. 5. Conclusions
The automatic extraction method of the left ventricular inner contour on contrast-enhanced ultrafast cine MR images is accurate and reproducible. Left ventricular volumes and the ejection fraction are rapidly and accurately evaluated by the biplane modified Simpson's method. The volume curve of the left ventricle is easily and quickly obtained using the monoplane area-length
131
method based on automatic extraction of the left ventricular inner contour. References [1] Hasse A. Snapshot FLASH MRI: application to TI, T2, and chemical-shift imaging. Magn Reson Med 1990; 13: 77-89. [2l Atkins DJ, Burstein D, Edelman RR. First-pass cardiac pcrfusion: evaluation with ultrafast MR imaging. Radiology 1990; 174: 757-762. [3] Fram J, Merboldt KD, Bruhn H, Gyngell ML, Hanicke W, Chien D. 0.3 second FLASH MRI of the human heart. Magn Reson Med 1990; 13: 150-157. [4] Matthaei D, Hasse A, Henrich D, Duhmke E. Cardiac and vascular imaging with an MR snapshot technique. Radiology 1990; 177: 527-532. [5] Holsinger AE, Wright RC, Riederer SJ, Farzaneh F, Grimm RC, Maier JK. Real-time interactive magnetic resonance imaging. Magn Reson Med 1990; 14: 547-553. [6] Mansfield P. Multiplanar image formation using NMR spinechoes. J Phys C Solid State Phys 1977; 10: L55-L58. [7] Rosen BR, Bclliveau JW, Vevea JM, Brady TJ. Perfusion imaging with NMR contrast agents. Magn Reson Med 1990; 14: 249-265. [8] Worthington BS, Mansfield P. The clinical application of echo planar imaging in neuroradiology. Neuroradiology 1990; 32: 367-370. [9] Edelman RR, Wailner B, Singer A, Atkinson D J, Saini S. Segmented turboFLASH: method for breath-hold MR imaging of the liver with flexible contrast. Radiology 1990; 177: 515-512. ll0] Atkinson DJ, Edelman RR. Cineangiography of the heart in a single breath hold with a segmented turboFLASH sequence. Radiology 1991; 178: 357-360. [11] Matsumura K, Nakase E, Haiyama T, Takeo K, Shimizu K, Yamasaki K, Kohno K. Determination of cardiac ejection fraction and left ventricular volume: contrast-enhanced ultrafast cine-MR imaging vs. IV digital subtraction ventriculography. AJR 1993; 160: 979-985. [12] Chapman CB, Baker O, Reynolds J. Use of biplane cinefluorography for measurement of ventricular volume. Circulation 1958; 18: ! 105-1 ! 17. [13] Dodge HT, Sandier H, Baxley WA. Usefulness and limitations of radiographic methods for determining left ventricular volume. Am J Cardiol 1966; 18: 10-24. [141 Matsumura K, Nakase E, Kawai I, Saito T. Clinical evaluation of cardiac intravenous digital subtraction angiographyfunctional analysis of the four cardiac chambers. Med Rev 1992; 41: 1-13. [15] Sechtem U, Pflugfelder PW, White RD. Cine MR imaging: potential for the evaluation of cardiovascular function. A JR 1987; 148: 239-246. [16] Higgins CB, Holt W, Pflugfelder P, Sechtem U. Functional evaluation of the heart with magnetic resonance imaging. Magn Resort Med 1988; 6: 121-139. [17] Buser PT, Auffermann W, Holt WW. Noninvasive evaluation of global left ventricular function with the use of cine nuclear magnetic resonance. J Am Coil Cardiol 1989; 13: 1294-1300. [18] Cranney GB, Lotan CS, Dean L, Baxley W, Bouchard A, Pohost GM. Left ventricular volume measurement using cardiac axis NMR imaging-validation by calibrated ventricular angiography. Circulation 1990; 82: 154-163. [19] Semelka RC, Tomei E, Wagner S. Normal left ventricular dimensions and functions: interstudy reproducibility of measurements with cine MR. Radiology 1990; 174: 763-768.
132
K. Matsumura et al./ European Journal of Radiology 20 (1995) 126-132
[20] Benjelloun H, Cranney GB, Kirk KA. Interstudy reproducibility of biplane cine nuclear magnetic resonance measurements of left ventricular function. Am J Cardiol 1991; 67: 1413-1420. [211 Teichholz LE, Kreulen T, Herman MW. Problems in echocardiographic volume determinations: echocardiographicangiographic correlations in the presence or absence of asynergy. Am J Cardiol 1976; 37:7-1 I.
[22] van der Gecst R J, Jansen E, Builer VGM, Reiber JHC. Automated detection of left ventricular epi- and endocardial contours in short-axis MR images. Comp Cardiol 1994; 33-36. [23] Rackely CE. Quantitative evaluation of left ventricular function by radiographic techniques. Circulation 1976; 54: 862-879.