h4agneric Resonance Imaging. Vol. 9, PP. 435-447. 1991 Printed in the USA. All rights reserved. 0731)-725X/91 $3.00 + .oo Copyright CC1 1991 Pergamon Prec\ plc l Original Contribution OPTIMIZED PULSE SEQUENCES FOR MAGNETIC RESONANCE MEASUREMENT OF AORTIC CROSS SECTIONAL AREAS MICHAEL H. BUONOCORE AND HUGO BOGREN Division of Diagnostic Radiology, UC Davis Medical Center, Sacramento, California 95817, USA This study was done to improve the ability of magnetic resonance (MR) imaging to provide clear cross-sectional images of the ascending and descending aorta in diastole. The study was motivated by interest in measuring the regional compliance of the ascending aorta, which requires determination of the change in cross sectional area of the vessel between systole and diastole. In diastolic images, residual signal from slow flowing blood and flow artifact consistently obscured the inner boundary of the aortic wall and precluded tracing and measurement of the cross sectional area. We concluded that cross sectional area measurement of the ascending aorta was impos- sible on our system using standard spin echo sequences. To improve wall delineation in diastolic images, SAT pulses were optimized with respect to pulse timing, slice thickness, and gap. Optimized SAT pulses greatly improved the delineation of the vessel wall by removing unwanted signal from flowing spins. Measurement precision was vastly improved by running two scans with and without flow compensation, and correlating visually and ruumerically the area measurements from each. We established that each image should be measured by two independent ob- servers and traced three times by each. Using these procedures, diastolic cross-sectional areas of the mid-ascend- ing aorta could be measured with a precision of 2.5%, and the change of cross-sectional area between systole and diastole could be measured with a precision of 10.8%. These measurements were precise enough to detect CAD patients with low aortic compliance from the age-matched controls previously reported in one study. The test based on cross sectional area measurement, with a false positive detection rate of 5%, had a false negative rate of 58%. Compliance measurements by MR at 1.5 T could become clinically useful if normal and abnormal populations are sufficiently separated. Keywords: Magnetic resonance (MR), pulse sequences: Magnetic resonance (MR), experimental; Aorta, MR studies. INTRODUCTION This study was done to improve the ability of mag- netic resonance (MR) imaging at 1.5 T to provide clear cross-sectional images of the ascending and descend- ing aorta in diastole. This study was motivated by the results of Mohaiddin et al., ’ who measured aortic compliance using MR at 0.5 T in 70 healthy volun- teers, 13 athletes and 17 patients with coronary artery disease (CAD), and found that compliance in the CAD patients was significantly lower than in the age matched controls. Measurement of regional compli- ance requires determination of the change in cross- sectional area between systole and diastole. In MR, this change can be determined by subtraction of inde- pendent systolic and diastolic area measurements. Using standard spin-echo pulse sequences, systolic im- ages provided clear delineation of the vessel wall and were easy to measure. Unfortunately, the diastolic im- ages could not be measured because residual signal from slow flowing blood and flow artifact consistently obscured the inner boundary of the aortic wall. The inner boundary had a variable appearance and differ- ent observers could not agree on the precise location of it based on the gray scale transitions in the images. We concluded that the change in cross sectional area of the ascending and descending aorta was impossible to measure using standard spin-echo sequences. To try to improve wall delineation in diastolic im- ages and make measurement feasible, the SAT pulses of otherwise standard spin-echo pulse sequences were optimized with respect to pulse timing, slice thickness, RECEIVED IO/l l/90; ACCEPTED 2/12/91. Address correspondence to Michael H. Buonocore, Di- Acknowledgment- This work was supported in part by vision of Diagnostic Radiology, TICON-II, UC Davis Med- a grant from the Whitaker Foundation. ical Center, Sacramento, CA 95817. 435
Transcript
h4agneric Resonance Imaging. Vol. 9, PP. 435-447. 1991 Printed in the USA. All rights reserved.
OPTIMIZED PULSE SEQUENCES FOR MAGNETIC RESONANCE MEASUREMENT OF AORTIC CROSS SECTIONAL AREAS
MICHAEL H. BUONOCORE AND HUGO BOGREN
Division of Diagnostic Radiology, UC Davis Medical Center, Sacramento, California 95817, USA
This study was done to improve the ability of magnetic resonance (MR) imaging to provide clear cross-sectional images of the ascending and descending aorta in diastole. The study was motivated by interest in measuring the regional compliance of the ascending aorta, which requires determination of the change in cross sectional area of the vessel between systole and diastole. In diastolic images, residual signal from slow flowing blood and flow artifact consistently obscured the inner boundary of the aortic wall and precluded tracing and measurement of the cross sectional area. We concluded that cross sectional area measurement of the ascending aorta was impos- sible on our system using standard spin echo sequences. To improve wall delineation in diastolic images, SAT pulses were optimized with respect to pulse timing, slice thickness, and gap. Optimized SAT pulses greatly improved the delineation of the vessel wall by removing unwanted signal from flowing spins. Measurement precision was vastly improved by running two scans with and without flow compensation, and correlating visually and ruumerically the area measurements from each. We established that each image should be measured by two independent ob- servers and traced three times by each. Using these procedures, diastolic cross-sectional areas of the mid-ascend- ing aorta could be measured with a precision of 2.5%, and the change of cross-sectional area between systole and diastole could be measured with a precision of 10.8%. These measurements were precise enough to detect CAD patients with low aortic compliance from the age-matched controls previously reported in one study. The test based on cross sectional area measurement, with a false positive detection rate of 5%, had a false negative rate of 58%. Compliance measurements by MR at 1.5 T could become clinically useful if normal and abnormal populations are sufficiently separated.
Keywords: Magnetic resonance (MR), pulse sequences: Magnetic resonance (MR), experimental; Aorta, MR studies.
INTRODUCTION
This study was done to improve the ability of mag- netic resonance (MR) imaging at 1.5 T to provide clear cross-sectional images of the ascending and descend- ing aorta in diastole. This study was motivated by the results of Mohaiddin et al., ’ who measured aortic compliance using MR at 0.5 T in 70 healthy volun- teers, 13 athletes and 17 patients with coronary artery disease (CAD), and found that compliance in the CAD patients was significantly lower than in the age matched controls. Measurement of regional compli- ance requires determination of the change in cross- sectional area between systole and diastole. In MR, this change can be determined by subtraction of inde- pendent systolic and diastolic area measurements.
Using standard spin-echo pulse sequences, systolic im- ages provided clear delineation of the vessel wall and were easy to measure. Unfortunately, the diastolic im- ages could not be measured because residual signal from slow flowing blood and flow artifact consistently obscured the inner boundary of the aortic wall. The inner boundary had a variable appearance and differ- ent observers could not agree on the precise location of it based on the gray scale transitions in the images. We concluded that the change in cross sectional area of the ascending and descending aorta was impossible to measure using standard spin-echo sequences.
To try to improve wall delineation in diastolic im- ages and make measurement feasible, the SAT pulses of otherwise standard spin-echo pulse sequences were optimized with respect to pulse timing, slice thickness,
RECEIVED IO/l l/90; ACCEPTED 2/12/91. Address correspondence to Michael H. Buonocore, Di- Acknowledgment- This work was supported in part by vision of Diagnostic Radiology, TICON-II, UC Davis Med-
a grant from the Whitaker Foundation. ical Center, Sacramento, CA 95817.
435
436 Magnetic Resonance Imaging 0 Volume 9, Number 3, 1991
and gap. Detailed instructions for manually tracing the vessel regions of interest were developed to stan- dardize the process among different observers, and these instructions were based on accepted MR physical principles. Finally, statistical analysis was performed to determine the precision of the cross sectional area measurements.
METHODS
Pulse Sequences for Cross-Sectional Area Measurement
All subjects were scanned on a GE. Signa 1.5 Tesla MR System using spin echo and gradient recalled echo pulse sequences. Four different spin-echo sequences and one gradient recalled echo sequence were used with the parameters listed in Table 1. The S/I level of the axial slice was determined from the axial images of a localizer series. The chosen level clearly showed the bifurcation of the main pulmonary artery, or the right pulmonary artery outflow. Twenty subjects were scanned using standard MEMP sequences to recognize and characterize the diastolic image problem. Six sub- jects were scanned with the full repertoire of standard spin-echo, optimized spin-echo, and gradient-recalled echo sequences for the generation of the statistical re- sults. Appendix 1 describes the calculation used to de- termine what spatial resolution was needed to measure the cross sectional area change. We calculated the ex- pected standard area of an area measurement based solely on the error due to the finite pixel size of the MR image, and found it to be small relative to the to- tal measurement errors. The 128 x 256 matrix size with 24 cm FOV is satisfactory, in that the other sources of error due to image characteristics of the
Table 1. MR scan parameters common to all pulse sequences
Gradient Parameter RF spin-echo recalled echo
Field of view (FOV) 24 cm 24 cm
Matrix size 128 x 256 128 x 256 Pixel size 1.875 x 0.938 mm 1.875 x 0.973 mm Pixel area 1.76 mm2 1.76 mm2 Slice thickness 5mm 5mm Excitations (NEX) 2 2 Repetition time
vessel wall overwhelm the blurring of the wall due to pixel size.
Axial multi-slice spin echo (MEMP) scans were set up so that at a 4-msec trigger delay the diastolic image at the level of interest would be obtained first in the multi-slice pass. Late diastolic images 80 msec to 0 msec before the EKG trigger could not be reliably ob- tained because of normal variations in the R-R inter- val. The scan was set up for subsequent slices to be obtained inferiorly, into the heart chambers. This al- lowed the RF pulses to provide incidental saturation of blood signal. A second scan was required to get the systolic image (at 250 msec trigger delay) at the level of interest. In subsequent experiments, single-slice, double-phase spin-echo imaging was used to acquire images only at the slice of interest. The advantages of this strategy was the ability to acquire both systolic and diastolic images in a single 4-min scan, and the ability to use optimized pre-saturation (SAT) pulses to suppress flow signal and artifacts.2’3 The pulse se- quences used in these experiments were modified to al- low two SAT pulses to be played out in the slice select direction. Variable rate RF pulses4 were used to make contiguous SAT slices, i.e., slices with sharper bound- aries and smaller gap compared to the SAT slices made by standard RF pulses. Flow compensated gradients were also used in the slice select and the frequency en- code directions. The spin-echo pulse sequence using flow compensated gradients was called FCMEMP (flow compensated MEMP), and the identical sequence without flow compensation was called CSMEMP (con- tiguous slice MEMP). Another pulse sequence devel- oped for this study, referred to as TEMP, employed transparent pulses for flow signal suppression. The theory and use of transparent pulses is described else- where.’ The gradient-recalled echo pulse sequence we used, referred to as VINNIE,6 provided both ampli- tude and phase images. Velocity data were obtained in the phase images and used in the development of the flow model (described below). The amplitude images were CINE images’ in which the blood signal was bright relative to the surrounding vessel wall.
Specification of Presaturation (SAT) Pulses SAT pulses were defined by the thickness of the sat-
uration (SAT) slice, the spatial gap between the image and SAT slice, and the timing delay between the SAT pulse and the 90-degree data acquisition pulse. These parameters determined the effectiveness of the SAT pulse to suppress the signal from blood flowing into the slice. The standard SAT pulses made an 80-mm- thick slice with a 30-mm gap, and were played out im- mediately after data acquisition. Optimized SAT pulse parameters used in the study were classified into three
Optimized pulse sequences 0 M.H. BUONOCORE AND H. BOCREN
Table 2. SAT pulse groups and flow signal suppression capability. Percent flow suppression and maximal achievable percent flow suppression are determined from Fig. 4
431
Maximum Time Peak achievable
SAT delay Thickness GAP velocity % flow % flow pulse (msec) (mm) (mm) (cm/set) suppression suppression
Close #l 180 10 2 8.0 56 60 Close #2 360 30 6 6.0 53 54
Far #l 250 60 12 28.0 68 77 Far #2 450 60 12 16.0 53 70
groups: “close,” “midrange,” and “far.” Table 2 lists the SAT pulse parameters used in each group. Note that two SAT pulses, to be played out at different times in the interval between the systolic and the sub- sequent diastolic data acquisition, were defined in each group. These groups and pulse parameters were selected based on a mathematical model of the flow and from experimental trials. Appendix 2 discusses the flow model in detail.
Instructions Used to Trace Vessel Boundaries Table 3 gives instructions for tracing the inner
boundaries of the vessel walls. Because CSMEMP (nonflow-compensated) and FCMEMP (flow-compen- sated) had different effects on the spins at the inner boundary, their images required different instructions for tracing the inner boundary. The different instruc- tions were based on different physical explanations (see Discussion) for why the images appeared as they did at the vessel wall. The instructions served to stan- dardize the tracing process among different observers doing the tracing. The tracing process included setting the width and level at fixed settings, so that the gray scale presented to each observer was the same. Be- fore actual tracing was done, the vessel wall inner boundaries of the two images needed to be roughly established visually using elliptical ROIs placed on the images. By interactively positioning and switching back and forth between the images, a single elliptical ROI provided rapid visual comparison and confirmation that traces made along specific lines of signal intensity were going to result in similar area measurements.
Statistical Analysis A four-level nested analysis of variance (ANOVA)
was used to determine the errors in diastolic and sys- tolic area measurements of both the ascending and
descending aorta.* Errors and biases in the measure- ments were identified by using two observers who traced the vessels independently, and having these ob- servers make three independent tracings at each of two sessions (two months apart). Using the quantitative re- sults from the ANOVA study, a protocol involving multiple tracings was devised for measuring the cross- sectional area change. Finally, as a demonstration, the devised protocol was applied to the detection of low cross-sectional area changes that had been previously measured’ in patients with coronary artery disease. The false positive and false negative rates for distin-
Table 3. Region of interest (ROI) tracing instructions
Name Instructions
“Smaller ROI” Inside edge of traced line follows the inner boundary of vessel wall, and blotches of signal in contact with the boundary are cut through in order to maintain local smoothness. The traced line is offset l/2 pixel width inside of this smoothed boundary. The inner boundary is defined at the middle of the gray scale transition from the dark lumen to the brighter wall.
“Larger ROI” Same as “smaller ROI”, except that the traced line is offset l/2 pixel width outside of the smoothed boundary.
“Alternate ROI” Inside edge of traced line follows alter- nate transitions of signal intensity, using either the “Smaller ROI” or “Larger ROI” instructions. Alternate ROI’s are those that are not the pri- mary candidate for the vessel wall as determined by an elliptical ROI survey.
438 Magnetic Resonance Imaging 0 Volume 9, Number 3, 1991
guishing these patients from an age-matched group of normal subjects were determined.
RESULTS
Initial Studies Using Nonoptimized Sequences Standard MEMP imaging did not produce images
in which area measurement could be reliably obtained. Typical images are shown in Fig. 1. Contiguous slice MEMP using standard SAT pulses (referred to as stan- dard CSMEMP) generated diastolic images of varying appearance which we have assigned into two broad categories. Representative images are shown in Fig. 2. Neither category of images were of sufficient quality to be useful for cross-sectional area measurements. About 50% of normal subjects presented in the first category, with bright walls and bright and variable in-
traluminal signal. Blotches of high signal from blood typically obscured the inner boundary of the vessel wall. The other 50% presented in the second category, with a low intensity vessel wall and similar low inten- sity intraluminal signal and artifacts.
Studies Using Optimized Pulse Sequences CSMEMP images using the “midrange” SAT pulses
(referred to as optimized CSMEMP) had a dark lumen and gray to bright vessel wall. These images were sim- ilar to the second category of standard CSMEMP im- ages. However, the wall was better defined, and the number and strength of alternate boundaries was re- duced. FCMEMP images with the “midrange” SAT pulses (referred to as optimized FCMEMP) had a gray to white lumen, and a brighter vessel wall that was thicker than that obtained with CSMEMP, and rela-
(A)
Fig. 1. Images from multislice multiecho (MEMP) sequence-trigger delay = 4 msec. Images from the MEMP pulse sequence using standard SAT pulses (A) the first image taken at the level of interest (B), and a schematic (C) to identify the vessels. Note that the systolic images of the multi-slice sequence have well delineated vessel and chamber boundaries. Only the diastolic images, especially the first of the series, has poorly delineated boundaries. (Figure continued on facing page.)
Optimized pulse sequences 0 M.H. BUONOCORE AND H. BOGREN 439
03
Fig. 1 continued.
0
tively easy to trace. The SAT pulses reduced the in- tensity of the lumen thus making a larger intensity transition from lumen to wall. Representative images for optimized CSMEMP and FCMEMP are shown in Fig. 3. CINE images were similar to FCMEMP in that the lumen was bright, but in CINE the wall was less bright than the lumen, and the transition between the bright lumen and gray wall was broader. CINE images were not used in the recommended protocol and are not shown. TMEMP images were similar to the 2nd category of CSMEMP images, but had darker lumen and thinner vessel wall. TMEMP images were also not useful for cross-sectional area measurements, and are not shown.
Calculating the Optimal SAT Parameters Figure 4 shows the SAT pulse parameters and flow
signal suppression capability for peak flows of 6.0, 8.0, 12.0, 16.0, 20.0, and 28.0 cm/set. These graphs were used to estimate the flow signal suppression that
would be achieved with the different SAT pulse groups. Using graph (A), the SAT time delay and thickness of any particular SAT pulse was set along the x and y axes, respectively, to determine the peak velocity. In graph (B), the SAT time delay was used with the peak velocity found from graph (A) to determine the per- cent flow signal suppression achieved. Table 2 shows the percent suppression of flow signal expected from each pulse of each SAT pulse parameter group. While theoretical SAT pulse performance could be computed from the model, experiments were needed to take into consideration unmodeled aspects of the problem. For example, the “close” group was not useful at all be- cause aortic root motion led to suppression of the vessel wall in addition to the flowing blood. The flow model indicated that the “midrange” group would be slightly better than the “far” group, and the experi- ments confirmed this. The midrange group most often gave the greatest improvement in delineation of the vessel wall and elimination of alternate boundaries.
440 Magnetic Resonance Imaging l Volume 9, Number 3, 1991
(W
Fig. 2. Example of (A) first and (B) second category of image appearance with standard CSMEMP. These images illustrate the two typical appearances of diastolic images using a single slice, double phase spin-echo pulse sequence without flow com- pensation or optimized SAT pulses.
Tracing the Inner Boundary of the Vessel Wall The best quantitative agreement was between the
“smaller ROI” instructions aDplied to optimized
CSMEMP images and the “larger ROY instructions applied to optimized FCMEMP images. Figure 3 presents images and ROIs showing this agreement. Images (C) and (D) illustrate the “smaller ROY and “larger ROI” instructions, respectively. The inside edge of the traced line defines the region of interest, because the image processor on the Signa system computes the ROI area as those pixels enclosed by the trace. The pixels on the trace are not part of the ROI. In all in- structions, the boundary was defined as the midpoint of the transition from the lumen to the brighter wall. In image (C), the inside edge was placed $ pixel width inside the boundary, so that a 4 pixel wide strip of low signal intensity was therefore excluded from the vessel lumen. In Image (D), the inside edge was placed 4 pixel width outside of the inner boundary. In this way, a f pixel wide strip of bright signal intensity was
included as part of the vessel lumen. The ROIs in (B) and (D), although traced on entirely different images and using different instructions, agreed in area to within 3%. Qualitative match of the vessel boundaries with an elliptical ROI was done prior to actual tracing to remove consideration of alternate boundaries, to set a definite path through ambiguities in the images, and to determine the precise tracing needed to yield consistent measurements.
Statistical Analysis The statistical analysis of the scan and measure-
ment protocol is presented in Tables 4 through 6. The ascending aorta cross-sectional area change between systole and diastole could be measured with a precision of 10.8% using the protocol. A demonstration of the use of the protocol is given showing that differences between the area changes reported in CAD patients and age-matched normal subjects can be distinguished. The false negative rate is higher than most accepted
09 (D)
Fig. 3. (A) Optimized CSMEMP image, (B) optimized FCMEMP image, (C) optimized CSMEMP imaged traced with the “smaller ROI” instructions, and (D) optimized FCMEMP image traced with the “larger ROI” instructions. Images (C) and (D) illustrate the different instructions used for tracing the inner boundary for each pulse sequence. Image (C) shows image (A) traced according to the “smaller ROI” instructions. Image (D) shows image (B) with the ROI traced according to the “larger ROI” instructions.
70
60
S A
T 50
S I i c e 40
T h i e ; 30
e $
s s ,”
m
10 0
iz
/ /
0 20
0 40
0 60
0 60
0 lo
o0
1200
14
00
SAT
Tim
e D
elay
- m
sec
l
vmax
= 6.0
0
vmax
= 8
.0
+ vr
nax
= 12
.0
0 vm
ax =
16.
0 -A
vm
ax =
20.
0 *
vmax
=
26.0
904
800
700
S A
T
600
T I m
e 50
0 D
e I a 40
0 Y m
30
0 s e C
200
100 0
-!-
20%
30%
4o%
6o%
6O%
Perc
ent
Flow
Sig
nal
Supp
ress
ed
l
vmax
= 6
.0 q v
max
= 6
.0 l v
max
- 1
2.0
0 vm
ax =
16.
0 *
vmax
= 2
0.0
&x
vmax
= 2
6.0
(B)
(4
Fig.
4.
Fl
ow m
odel
de
term
inat
ion
of f
low
sig
nal
supp
ress
ion.
T
he f
low
mod
el
dete
rmin
es
the
achi
evab
le
perc
ent
flow
sig
nal
supp
ress
ion
base
d on
the
im
age
and
SAT
sl
ice
thic
knes
s,
SAT
gap
, an
d SA
T t
imin
g de
lay.
O
n gr
aph
(A),
SA
T s
lice
thic
knes
s an
d tim
e de
lay
pair
is
fou
nd
alon
g a
line
to d
eter
min
e th
e pe
ak v
eloc
ity
that
th
e pa
ir i
s op
timiz
ed
for.
On
grap
h (B
), t
he s
elec
ted
SAT
tim
e de
lay
and
peak
vel
ocity
fr
om g
raph
(A
) de
term
ines
th
e pe
rcen
t fl
ow s
igna
l su
ppre
ssio
n fo
r th
ose
para
met
ers.
Optimized pulse sequences 0 M.H. BUON~CORE AND H. BOGREN 443
tz.7 = p < 0.90; no notation, p = 0.90-<0.95; *p = 0.95-<0.99; **p = 0.9940.999; ***p > 0.999.
tests for CAD in nuclear medicine (e.g., 58% associ- ated with a 5% false positive rate), but the only way that the false negative would be decreased is if the pa- tient and age-matched normal populations turned out to be better separated. With respect to the CAD pa- tient data analysis, it is important to emphasize that we are not making a statement as to the relevance of area changes for detecting coronary artery disease. Rather, we are only confirming that the MR protocols developed in this paper would be able to distinguish the low cross-sectional area changes typical of this group of CAD patients from the group of age-matched normal subjects.
The F-tests in the 4-Level Nested ANOVA show that most of the variance was due to the true differ-
ences in cross sectional areas among the different sub- jects. Performing the measurement at two different sessions (which were separated in time by two months) revealed a large component of variance, indicating that the observers interpreted the tracing instructions somewhat differently in each session. The added vari- ance component due to observers was not significant (ns) in any of the measurements, indicating that there was no bias of one observer relative to the other in the interpretation of the tracing instructions. The added variance due to the pulse sequences was significant (p < 0.05) for the diastolic descending aorta only. The fact that the added variance was not significant for the ascending aorta indicates that the different tracing instructions applied to the pulse sequence’s im-
Table 5. Normal subject area measurement statistics
Units in cm*
Ascending Aorta
Diastole Systole
Descending Aorta
Diastole Systole
Average C-S area Standard deviation of area Standard error of single measurement Standard error of protocol measurement Standard error as % of average
444 Magnetic Resonance Imaging 0 Volume 9, Number 3, 1991
Table 6. Ascending aorta cross-sectional area statistics for coronary artery disease (CAD) patients and age-matched normal subjects
Units in cm2 df
CAD patients 16 Normal subjects 16 Difference 16
Average C-S area change
0.464 0.704 0.240
SD Std. error of SD of of C-S area protocol
population change measurement
0.176 0.071 0.162 0.419 0.386 0.162 0.454
(A)
Identification of subjects with low cross-sectional areas Tests based on Protocol Measurements*
Prob. false positive Prob. false negative
25.0% 40.2%
10.0% 50.9%
5.0% 57.6%
1 .O% 70.4%
0.1% 83.7%
*False positive and negative pairs arranged vertically. (B)
ages are consistent for this vessel. Table 5(A) shows that a single measurement (i.e., one tracing by one ob- server on one CSMEMP or FCMEMP image at one session) had a standard error of 0.256 cm* for dias- tole and 0.186 cm* for systole. If the protocol was used, the reported measurement was then an average of 12 independent measurements (i.e., three repetitions of the tracing by two observers on both CSMEMP and FCMEMP images), and the standard error was reduced to 0.116 cm* for diastole and 0.112 cm2 for systole. The protocol reduced the measurement stan- dard error to 2.45% and 1.79% in diastole and sys- tole, respectively. Table 5(B) shows that the single measurement of the area change had a standard error of 0.317 cm*, which is more than 20% of the mean area change. Using the protocol, the standard error was reduced to 0.162 cm2, or 10.8% of the mean area change.
Table 6(A) shows data from the previous MR mea- surements of aortic cross sectional area in normal sub- jects and CAD patients. The mean area change in CAD patients was 0.464 cm*, and in age-matched normal subjects was 0.704 cm*. The estimates of the standard error of these estimates of the population means were 0.176 cm* and 0.419 cm*, respectively. We extracted the standard deviation of true differ- ences in cross sectional area change among subjects from the standard error due to the measurement pro- cedure. Table 6(B) shows the false positive and nega- tive error rate pairs using the protocol for detection of the low aortic cross sectional area change characteris- tic of CAD. The hypothesis being tested was that the individual is from the abnormal population.
Recommended Pulse Sequence and A4easurement Protocol
The recommended protocol for ascending aorta cross-sectional area measurement consists of three scans. The first should be a multislice standard MEMP localizer scan to find a level of the mid ascending aorta at the pulmonary artery bifurcation. This should be followed by an axial single slice, double phase FCMEMP scan using the “midrange” SAT parameter group. The last scan should be an axial single slice, double phase CSMEMP scan also using the “mid- range” group. The protocol for measuring the vessel cross sectional areas consists of tracing ROI’s in both the CSMEMP and FCMEMP images. The “smaller ROI” instructions are recommended for the CSMEMP images, and the “larger ROI” instructions for the FCMEMP images. An elliptical ROI survey should be used to establish the exact location of the vessel boundaries in the FCMEMP and CSMEMP images that give the best qualitative match. Each image should be traced three times by two observers and all the measurements averaged. Although not required (and not used in the statistical determination of the measurement standard error), tracing in two separate sessions is advantageous to further improve precision.
DISCUSSION
The SAT pulse parameters we used in the flow model do not fall on the absolute maximum flow sup- pression for the particular peak velocity selected be- cause the groups were originally devised using a model that did not include the T, recovery effects. Without
Optimized pulse sequences 0 M.H. BUONOCORE AND H. BOOREN 445
r, recovery effects, the model showed that the percent flow suppression could always be increased simply by having a wider SAT slice thickness with a longer SAT pulse timing delay. This result of the original model was clearly unrealistic so T, effects were put in. How- ever, it was deemed unnecessary to redo the study using a slightly different set of SAT pulse parameters which were slightly more optimal according to the model. True velocity distributions in vessels are much more complicated than laminar and vary from subject to subject, so it was not reasonable to expect more from this numerical model than simple confirmation that the pulse parameters were in the correct range for the typical diastolic flows.
Separation of image intensity from slow flowing blood along the vessel wall inner boundary from the wall itself was the most subtle and difficult ambiguity to resolve. In optimized CSMEMP images, signal loss in pixels at the inner boundary of the wall was pre- sumed to occur from phase differences between blood and wall spins. But, exactly how many pixels were in- volved in this effect on the boundary was unknown. The “smaller ROI” instructions assumed that the effect occurred in one pixel thickness at the inner boundary, so that the pixels at the boundary were dark. In opti- mized FCMEMP images, signal enhancement along the inner boundary of the wall was presumed to occur from phase refocusing of blood and wall spins. The “larger ROI” measurement instructions assumed that the effect occurred in one pixel thickness at the inner boundary, so that the pixels at the boundary were bright.
Aortic compliance measurement by MR was first carried out on a 0.5 T superconducting system.’ Dia- stolic and systolic images were obtained in two separate single-slice, single-phase scans. These scans did not employ flow-compensation or presaturation pulses to reduce flow artifacts in diastolic images, yet measure- ment precision was adequate. At both field strengths, even with optimized SAT pulses, flow artifacts can re- main because blood flowing slowly enough can remain entirely within the imaged slice and be unaffected by
Vessel Boundary
Pixels of 128 x 256 Matrix ,
Vessel Boundary
Fig. Al. Area measurement error due to finite pixel size. A circular vessel was used. Tracing along the boundary maps out “wedges” representing portions of the total vessel area. The error in each wedge area due to pixel size was +i the area of one pixel, irrespective of the orientation of the pixel relative to the traced line. The error in each wedge was as- sumed to be uncorrelated with the errors in the other wedges. The sum over all the wedges determined the amount of error in the area measurement.
the SAT pulse. Especially at high field, a technique providing suppression regardless of blood velocity is needed. An inversion recovery pulse sequence, with TI set to null the signal from blood, has been reported to eliminate signal from blood while providing adequate tissue signal to noise’ on a 1.5 T system.
APPENDIX 1 ANALYSIS OF CROSS-SECTIONAL AREA
MEASUREMENT ERRORS DUE TO FINITE PIXEL SIZE
This section presents a simple calculation of the er- ror in cross-sectional area measurement due to the fi- nite pixel size of the image. The following discussion refers to Fig. Al. The vessel area was partitioned into independent contributions from wedges with lateral boundaries defined by the trace of the vessel wall. We
Table A.l. Standard errors of area measurement due to finite pixel size from 24 cm FOV. 128 x 256 matrix
Normal subjects areas
Mean Low High
Units in cm2.
Diastolic C-S area
4.80 4.00 7.80
Systolic C-S area
6.30 5.50 9.30
Diastolic standard error
1.40% 1.61% 0.97%
Systolic standard error
1.27% 1.14% 0.85%
Area change
1.50 1.50 1.50
Area change standard error
6.31% 6.56% 7.32%
446 Magnetic Resonance Imaging 0 Volume 9, Number 3, 1991
modelled the error of the vessel area as arising from an accumulated sum of the errors from each wedge area. The calculation proceeded by assuming that the error in positioning the cursor to set the lateral bound- ary of the wedge results in a random error of &h pixel area. The calculation then assumed that the er- rors from the wedges are uncorrelated, so the total er- ror of the vessel area was im Area of One PixeI, where N was the number of wedges. The number of wedges was determined by assuming a circular vessel. Table Al -shows the errors due to finite pixel size of the image being traced. The largest source of error was due to the limitations inherent to the image and the data acquisition. Because the spatial resolution due to these other factors dominates, it is likely that the 128 x 256 had no degrading effect on the measure- ment precision.
APPENDIX 2 FLOW MODEL FOR ESTIMATING
FLOW SIGNAL SUPPRESSION
The model was designed to provide a rough esti- maje of the SAT pulse parameters needed to obtain signal suppression of flowing spins in diastole. Mean diastolic flows in the ascending aorta were approxi- mately 5 cm/set. A T1 of 800 msec for blood was used. Three different sets of SAT pulse parameters were used to allow for a range of actual diastolic ve- locity profiles. The following discussion refers to Fig- ure A2. The model assumed a constant laminar flow profile. The imaged slice was 5 mm. The model re-
Vmax
Fig. A2. Flow model for computing percent flow signal sup- pression. Within the vessel boundary (thick lines), laminar flow partially replaces the spins within the imaged slice with spins from the SAT slice. The arrows indicate the distance traveled by the spin at the labeled velocity in the SAT delay time. Geometric construction was used to determine the per- cent replacement within the imaged slice of spins that have been affected by the SAT pulse. T, recovery of the slices was also included in the model.
quired input of the peak velocity (I&,), which under the laminar flow assumption was twice the mean ve- locity. The gap was set at 20% of the SAT slice thick- ness plus the image slice thickness, which by MR experimentation was found to be adequate to prevent the SAT pulse affecting spins within the imaged slice, as long as variable rate RF pulses were used. The SAT time delay was set so that the maximum volume of spins from the SAT slice moved into the imaged slice by the time of excitation. The model considered the time it takes for the leading edge of the SAT slice to move to the distant border of the imaged slice, and the time it takes for the spins at the distant border of the SAT slice to move into the imaged slice. Thus, as V,, is increased a thicker slice was used to suppress all spins able to move into the imaged slice, but then the resulting larger gap dictated a longer SAT time de- lay so that the slower velocity spins at the periphery and in the gap would be able to move through the im- aged slice. In the figure, Vi is the minimum velocity of spins for which a maximum percentage of the spins within the imaged slice had been suppressed by the SAT pulse. V, is the maximum velocity for which Oslo of the spins within the imaged slice had been sup- pressed by the SAT pulse. Finally, in the range VZ < u < Vi, a fraction of the spins in the imaged slice had been suppressed by the SAT pulse. This fraction var- ies from 0% for the spins with velocity Vz, to a max- imum percent for the spins at velocity Vi. The effect of Tl must also be included. The graphs were gener- ated by first computing the percent of spins in the im- aged slice that have been suppressed by the SAT pulse. This result would be the percent flow signal sup- pressed if no recovery of z magnetization occurred. Based on the computed SAT timing delay, this percen- tage of spins was adjusted downward to take into ac- count the recovery of z magnetization during the SAT timing delay. For example, if the SAT timing delay was 800 msec, equal to T, of blood, and the percen- tage of spins in the imaged slice that were affected by the SAT pulse was 90%, then the actual flow signal suppression would be computed as only 0.90 * 0.37 = 29.7%. The TI recovery prevented longer and longer SAT pulse delays leading to better and better flow sig- nal suppression.
REFERENCES
. Mohiaddin, R.H.; Underwood, S.R.; Bogren, H.G.; Fir- min, D.N.; Klipstein, R.H.; Rees, R.S.O.; Longmore, D.B. Regional aortic compliance studied by MRI: The ef- fects of age, training and coronary artery disease. Br. Heart J., 6290-96; 1989.
Optimized pulse sequences 0 M.H. BUONOCORE AND H. BOOREN 447
2. Ehman, R.L. Spatial presaturation: Comparison with other methods for reducing flow and motion artifacts, MR imaging of blood flow, SMRM Workshop, Philadel- phia, PA; March 13-14, 1990; p. 102.
3. Felmlee, J.P.; Ehman, R.L. Spatial presaturation: A method for suppressing flow artifacts and improving depiction of vascular anatomy in MR imaging, Radiology 164:559-564; 1987.
4. Clover, G.H.; Conolly, S.; Shimakawa, A. Variable rate selective RF pulses for time reversal and saturation, Book of Abstracts, SMRM Sixth Annual Meeting, New York: August 17-21, 1987:~. 159.
5. Buonocore, M.H. The analytic theory, optimization, and performance of transparent pulses, Book of Abstracts,
SMRM Eight Annual Meeting, Amsterdam, August 12- 18, 1989:~. 866.
6. Pelt, N.J.; Shimakawa, A.; Glover, G.H. Phase contrast tine MRI, Book of Abstracts, SMRM Eighth Annual Meeting, Amsterdam, August 12-18, 1989:~. 101.
7. Glover, G.H.; Pelt, N.J. CINE magnetic resonance im- aging. In: H.Y. Kressel (Ed), Magnetic Resonance An- nual 1988, New York: Raven Press; 1988:~. 299.
8. Sokel, R.R.; Rohlf, F.J. Biometry, 2nd ed. New York: Freeman and Co.; 1981.
9. Hinks, R.S. The elimination of signal from slow blood in MR of the heart and great vessels, Works-In-Progress, SMRM Eighth Annual Meeting, Amsterdam, August 12- 18, 1989:~. 1054.
