Magnetic Resonunce Imaging, Vol. 10. pp. 25-34, Printed in the USA. All rights reserved.
0730-725x/92 $5.00 + .oo 1992 Pergamon Press plc
1992
copyrigt0
l Original Contribution
MR ANGIOGRAPHY WITH PULSATILE FLOW R.G. DE GRAAFAND J.P. GROEN Philips Medical Systems, P.O. Box 10.000, Best, 5680 DA The Netherlands To achieve acceptable scan times, current multiple thin slice and 3D MR anglograpby (MRA) methods usually are based on continuous data acquisition, without ECG-synchronlxation. Tbe purpose of this work is to study consequences of pulsatlle blood flow for the 2D inilow method. Arterial blood flow and blood slgnai intensity versus cardiac phase were studied by a 2D phase based method with retrospective cardiac synchronlzatlon. Such studies were performed in different parts of the body and with different excitation flip angles. As expected, a clear relation between intensity enhancement and time dependent flow can be demonstrated. The raw data of these multiphase studies was used to simulate alternative htflow MRA data acqulsltion strategies to improve image quality, without the excessive increase in scan thne huplled by standard cardiac trlggerhtg. The alternatives investigated were data collection during part of the cardiac cycle and cardiac-ordered phase encoding. Simulation results indicate that the best results are obtained by a combination of both strategies. This method was implemented on Philips Gyroscan systems to compare it with standard nontrlggered 2D inflow in practical MRA studies. For highly pulsatile flow, much better MR angiograms were obtained in this way. Keywords: MRI; Angiography; Pulse sequences; Puisatile blood flow.
The early MR angiography (MRA) methods took into account the pulsatile nature of arterial blood flow by applying ECG-triggered acquisitions. l4 Recently, the emphasis has shifted to nontriggered MRA methods, which can provide 3D angiographic data in an acceptable scan time. The 2D and 3D inflow methods5-7 are both based on a short TR gradient echo sequence, which saturates proton spins in stationary tissue and highlights fresh unsaturated proton spins flowing into the imaging slice or volume. Generally, velocity-compensated sequences are used to make the phase of the MR signal insensitive to flow velocity. However, during parts of the cardiac cycle with a very low flow velocity, the amplitude of the signal from proton spins in blood will decrease due to saturation of the spins. Still, it has been found possible to obtain MR angiograms, for instance of the carotid and cerebral arteries, by the nontriggered inflow-based methods. Good MRA results have even been shown for the 2D and especially the 3D phase-based subtraction
method’ applied in a nontriggered mode, despite their necessary deviations from true velocity compensation. The best angiographic results by nontriggered methods are obtained for the head and neck. Other regions, for example, the legs, often produce relatively poor angiograms. From the arguments given above, the quality of arterial MR angiograms in different parts of the body is expected to be related to the pulsatility of arterial blood flow. The aim of this work is to further explore this influence, with emphasis on the 2D inflow method. First, our method to study pulsatile blood flow versus cardiac phase will be outlined. Results obtained in different parts of the body will be discussed. On the basis of this information, simulations of the standard nontriggered MRA methods and of modified acquisition schemes will be presented. Our first approach to improve MRA data is a gating strategy, in which data is acquired only during part of the cardiac cycle. During this gate, signals are measured for a number of phase encodings. To preserve a steady state, spin excitations without data acquisition are per-
RECEIVED 519191; ACCEPTED 817191. Address all correspondence to R.G. de Graaf, Philips Med-
ical Systems, P.O. Box 10.000, Best, 5680 DA The Netherlands.
INTRODUCTION
25
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
26
cardiac
cycle
A
B Fig. 1. Magnitude images of a slice through the abdominal aorta. TR = 30 msec, flip angle fW, slice thickness 4 mm. (A) Systolic phase. (B) Diastolic phase.
formed during the remainder of the cardiac cycle. This gating approach has a scan time penalty, determined by the gate width compared to the cardiac period. The second approach is an ordering strategy for phase encodings in relation to the cardiac cycle. The original linear phase-encoding order changes in this approach into interleaved linear k,-sweeps. This method is analogous to respiratory ordered phase encoding (ROPE)9 for respiratory artifact reduction. Very recently, Cho et al. published results on a similar cardiac phase encoding ordering approach for CSF flow artifact reduction. lo Finally, a combination of gating and ordering has been studied. Simulation of the various methods, using the raw data from our pulsatile flow studies, allows us to evaluate the proposed strategies without considering all practical implementation details. The combination of the gating and ordering approaches was implemented on an MR system for investigation of its expected qualities in practice.
caraiac cycle
B Fig. 2. Variation of the MR signal of blood in the internal carotid arteries through the cardiac cycle for excitation flip angles of 30 and 60”. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
PULSATILE FLOW INVESTIGATION A gradient-echo sequence with retrospective gating” has been applied to obtain angiographic data on a single slice for a large number of points in the cardiac cycle. Unlike cardiac triggering, this technique provides steady-state acquisition with the ability to reconstruct
MR angiography with pulsatile flow 0 R.G.
I
I
I
cardiac
0.0
1.0
J.P.
,
-101
I
cycle
DE GRAAF AND
0.0
A
27
GROEN
,
,
cardiac
cycle
/
j 1.0
A I --7---T-----j
I
I
~
I
I
_~- _.
30 deg
_’ . .
.
‘r’?
G ;cardiac 0.0
&A
_
1
1.0
wardiac 0.0
c&FL
1.0
B
Fig. 3. Variation of the MR signal of blood in the abdominal aorta through the cardiac cycle for excitation flip angles of 30 and 50”. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
Fig. 4. Variation of the MR signal of blood in the femoral arteries through the cardiac cycle for excitation flip angles of 30 and 60“. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
images for cardiac phases over the complete heartcycle. Thin slice images were acquired with different flip angles in several parts of the human body, such as neck, breast, abdomen, and extremities. The saturation effects of blood and static tissue were studied as a function of flip angle and cardiac phase. Interleaved with phase-en-
coding gradient values, two values of a flow sensitizing gradient were applied, and phase-difference images were calculated to quantify flow perpendicular to the slice. An example of systolic and diastolic magnitude images in the abdomen is given in Fig. 1. Flow curves over the cardiac cycle, obtained by the retrospectively
28
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992 40
0.0
I
I
I
I
30 deg ’ I cardiac
I
I cycle
I 1.0
A I
I. Fig. 5. arteries 30 and curves curves.
I
I
B
I
.
Variation of the MR signal of blood in the popliteal through the cardiac cycle for excitation flip angles of 60”. TR = 30 msec, slice thickness 4 mm. (A) Flow derived from phase difference data. (B) Magnitude
gated phase-based technique described above are given in Figs. 2(A)-5(A) for the carotids, the abdominal aorta, the femoral, and the popliteal arteries, respectively. The corresponding signal amplitude curves, obtained from the same regions of interest in the magnitude images, are given in Figs. 2(B)-5(B). Both the flow curves and the signal amplitude curves are given for excitation flip angles of 30 and 60”.
I -“Y “__
I
I
phase encoding
I
I
mtmber
Fig. 6. Simulation result of the standard nontriggered 2D inflow method for the femoral arteries. TR = 30 msec, flip angle 60“. slice thickness 4 mm. (A) Magnitude image. Large signal intensity variations versus heart phase give rise to ghost artifacts (arrow). (B) Signal variation as a function of phase encoding number.
DISCUSSION OF PULSATILE FLOW RESULTS As can be seen in Fig. 1, there is a considerable difference in image intensity for the abdominal aorta between systole and diastole. During systole, the high flow rate ensures effective inflow of fresh spins. As a result, a high signal intensity is observed for the abdominal aorta. Less efficient refreshment during diastole leads
MR
c
-KY
I
I
angiography with pulsatile flow 0 R.G.
I
I
Phase encoding
nu&er
+KY
DE GRAAF AND J.P. GROEN
-KY
29
Phase encoding
nuntber
+KY
B
6
Fig. 7. Simulation result of the ECG-gated 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. Prominent ghost artifacts are present (arrow). (B) Signal variation as a function of phase encoding number.
Fig. 8. Simulation result of the ECG ordered phase encoding 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. Vessels tend to be blurred (arrow) due to pulsatile flow, especially when larger excitation flip angles are used. (B) Signal variation as a function of phase encoding number.
to a considerably lower end-diastolic signal intensity of the aorta. The flow curves are more or less identical for both flip angles. This was expected because the flip angle
mainly influences signal amplitude. The correlation between the flow and the signal intensity curves can be explained by inflow refreshment. High flow results in an effective refreshment and thus a high signal intensity. For small flip angles, the increase in signal intensity is relatively smaller, but persists during more excitations after a flow peak. The pulsatility of flow depends on the anatomical region. For the carotid arteries (Fig. 2), the average in-
tensity over a vessel cross-section shows relatively small variations over the cardiac cycle. However, intensity inhomogeneities were observed during systole, which are probably due to turbulent flow. Stronger pulsatility of flow, resulting in stronger signal intensity variations, were found for the abdominal aorta and the femoral and popliteal arteries. Large veins such as the vena cava and the lower part of the jugular veins showed some pulsatility effects as well. Experiments described here were performed on healthy volunteers. Note, however, that vascular diseases may increase the pulsatility, resulting in more flow
30
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
artifacts for the nontriggered 2D inflow method. This may be the case for example with atherosclerotic disease,which decreases the flexibility of the vessel walls. In conclusion, signal intensity variations due to pulsatile blood flow are an important problem for inflow based MR angiography methods. In nontriggered applications, artifacts and loss of vascular contrast are due to occur. Phase-based angiography is less sensitive to saturation effects, because generally smaller flip angles are used compared to the inflow technique. However, the phase information is directly related to the velocity of blood, and therefore sensitive to pulsatile flow as well. For such methods, reverse flow will even reverse the contrast. For both types of methods, other effects related to the cardiac cycle, such as accelerated flow and movement of vessels, will add to the artifact level and the loss of vascular contrast.
SIMULATION OF MEA METHODS Raw data of the retrospectively gated scans described above together with the corresponding ECG timing data enabled us to simulate different acquisition strategies for MR angiography on pulsatile vessels. Simulations were performed for the standard 2D inflow method, and for the gating and ordering approaches described in the introduction. The retrospective ordering algorithm was adapted to provide images representative for the different acquisition strategies. Standard retrospective ordering combines data for one specific cardiac phase into an image. In the simulations of MRA methods, a cardiac timing variable was incremented with the repetition time for each subsequent spin excitation. Data corresponding to this cardiac timing variable was selected from the multiphase raw data set. For a nongated measurement, the timing variable was reset to zero each time the length of the cardiac cycle was exceeded. In simulations of the proposed gating method, the timing variable was reset to the beginning of the gate when it exceeded the end of the gate. To simulate ordering of phase encodings reiative to the cardiac cycle, the cardiac timing variable was incremented with the average heart beat interval divided by the total number of spin excitations. Thus, the range of phase encodings is mapped on the cardiac cycle. For the combination of gating and ordering of phase encodings, the cardiac timing variable was initially set to the start of the gate. For each phase-encoding or averaged signal it was incremented with the width of the gate divided by the total number of measurements, and the corresponding data was selected from the multiphase data set. Thus, a mapping of all phase-encoding values on the duration of the gate was achieved.
-KY
phase encoding mmd3er
+Ky
B Fig. 9. Simulation result of the gated sweep 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. (B) Signal variation as a function of phase encoding number.
SIMULATION RESULTS Results for the standard nontriggered 2D inflow method are shown in Fig. 6. A linear order of phase encodings was assumed. To achieve a high contrast for MRA, a data set acquired with a 60” flip angle was utilized for this simulation. Large signal intensity variations versus heart phase, as demonstrated in the retrospectively gated images, give rise to ghost artifacts (arrow) and a low vessel intensity in the nontriggered 2D inflow simulation. Ghost images of vessels occur due to periodical variations in signal intensity as a function of phase-encoding gradient value, as illustrated schematically in Fig. 6(B). The differences in flow pul-
MR angiography with pulsatile flow 0 R.G.
DE GRAAF AND J.P. GROEN
31
between (A) gated sweep and (B) standard nontriggered 2D inflow acquisition of the aorta bifurcation. Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 4 mm.
Fig. 10. A comparison
satility explain why nontriggered angiography of the carotids and the cerebral arteries gives reasonably good results, while artifacts and loss of signal are found for the lower aorta and the illiac arteries. The alternative acquisition strategies were examined on their ability to increase vascular contrast, reduce artifacts, and maintain sharp vessel edges. The discussions are based on the 2D inflow method, but similar arguments can be made for phase-based MBA methods. Simulations were carried out for different parts of the body to investigate the achievable improvements of alternative methods for different characteristics of pulsatile flow. The ECG gating method achieves a reduction of pulsatile flow effects at the cost of scan time. The width of the gate and its delay with respect to the R-wave were varied. This allowed adjustments of the gate in such a way that refreshment of blood is guaranteed for all acquired phase encoding steps. An example of the results is shown in Fig. 7(A). In this case, the gate was about 50% of the cardiac cycle starting at the rising edge of the systolic flow peak. An illustration of the remaining variations in signal intensity is given in Fig. 7(B). Compared to the nongated approach, improved arterial contrast is obtained. However, prominent ghost (arrow) artifacts are still present. The ghost artifacts are enhanced due to the fact that a fixed gate width results in an exactly periodical variation of signals versus phase-
encoding number. This should be compared to a nongated method, where ghosts are smeared to some extent by variations in heart rate. Unlike gating, an ordering approach does not introduce a time penalty. The acquisition order is changed in such a way that a single period of cardiac motion covers the complete range of phase-encodings. In practice, this will be achieved by multiple interlaced sweeps through the range of phase encoding values, synchronized with the cardiac cycle. A linear relationship between cardiac phase and phase-encoding number results, with an adjustable phase offset between the two. An example of the results is shown in Fig. 8. Ordering of the phase-encoding gradient relative to the cardiac cycle results in simulated images free of artifacts, although vessels tended to be blurred (arrow) for very pulsatile flow, especially when larger excitation flip angles are used. This result can be compared to similar observations of blurring for ROPE applied to respiratory motion. In our case, the lower phase-encoding numbers are acquired during the systolic part of the cardiac cycle, which results in improved contrast. However, as can be seen in Fig. 8(B), the higher phase-encoding numbers occur at the end-diastolic phase, where the slowly flowing blood is saturated. Thus, the expected image resolution for vessels is not fully achieved in the phase encoding direction. Cardiac gating with ordering of phase encodings can
32
Fig. 11. A comparison
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
between (A) gated sweep and (B) standard nontriggered
2D inflow acquisition of the popliteal trifurcation.
Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 3 mm. combine the advantages of both methods mentioned previously. In this case, a full sweep of phase-encoding values is performed for each gate. Signals for intermediate phase-encoding values are measured in subsequent cardiac cycles. The gate delay and width can be adjusted for optimal results, depending on the region of interest in the human body. The applied gate ensures enough refreshment of blood, while the ordering of phase encodings eliminates ghost artifacts caused by the remaining signal strength variations. An example of simulation results is shown in Fig. 9(A). Figure 9(B) illustrates the signal variation versus kYfor this method. This approach turns out to be very effective to obtain angiographic images free of artifacts
and with a good vessel definition and contrast. Practical implementation is facilitated by the fixed width of the gate, that is, in fixed phase-encoding sweeps which can be calculated in advance. This method, which will be referred to as Gated Sweep has been investigated in practice, as will be described in the next section. IMAGING RESULTS The Gated Sweep method was implemented on the Philips Gyroscan T5 and ACS systems (Philips Medical Systems, Best, The Netherlands) for comparison with the standard nontriggered 2D inflow method. Using an ECG signal, the Gated Sweep method acquires data dur-
MR
angiography withpulsatileflow0
R.G. DEGRAAFANDJ.P. GROEN
33
Fig. 12. A gated sweep 2D inflow acquisition of the aortic arch. Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 3 mm. Four overlapping partial projections are shown.
ing a gate in the cardiac cycle, while dummy spin excitations are performed in the remaining time. During each data acquisition gate, a sweep through the full range of phase-encoding values is performed. Intermediate phase-encoding values are applied in subsequent cardiac cycles. Depending on the anatomical region to be imaged, gate width and delaly from the R-wave can be adjusted. Typically, a gate width of 50% of the cardiac cycle is selected. Thus, the scan time is increased by a factor of two compared to standard nongated methods. In comparisons, this was compensated by using twice the number of measurements for the standard 2D inflow method. Results of Gated Sweep and standard 2D inflow are shown in Fig. 10 for the aorta bifurcation and in Fig. 11 for the popliteal trifurcation. The delay of the gate relative to the R-top of the ECG was adjusted according to the delay in onset of flow for the vessel of interest. For the aorta and the politeal trifurcation, delays of 150 msec and 250 msec were used, respectively. The gate width is selected to ensure sufficient signal intensity enhancement by refreshment of blood. For 2D inflow measurements, a width of 50% of the cardiac cycle was typically used. Both parameters are not very critical, and fixed values can probably be used for different patients. In case of aorta bifurcation (Fig. lo), the Gated Sweep method results in an increased intensity of the aorta and eliminatiaon of stripe patterns caused by pulsatile flow. Similarly, the arterial structures just below the knee (Fig. 11) are increased in intensity compared
to standard 2D inflow while veins with a rather constant flow are not affected. For MRA of the carotid arteries, the gate width could be adjusted to almost the full cardiac cycle. In that case, pulsatile flow artifacts are only suppressed by ordering of phase encodings and no significant loss of scan time occurs. For cerebral MRA, no significant improvements were observed. The aortic arch is a difficult area for nontriggered MRA methods, since movements of the arch will lead to artifacts. Figure 12 shows results obtained with the Gated Sweep method. In this case, the gate delay was 100 msec and the width 300 msec. Slices were acquired with a transverse orientation. The results of a partial projection clearly display the left and right common carotid arteries and the subclavian arteries. CONCLUSIONS Our study of flow pulsatility influences shows clear differences between different anatomical regions, with important consequences for MRA acquisition methods. Improved methods were developed for strongly pulsatile vessels. An important reduction in artifact level and improvement in vascular intensity has been demonstrated with a combination of cardiac gating and ordering of phase encodings. The gating approach introduces a time penalty of a factor of two, but is much more effective than signal averaging. Results on the Gated Sweep method presented here are preliminary because
34
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
clinical evaluation of this method is still to be performed. Choices for the gate width and gate delay need to be evaluated thoroughly in such a study. Similar techniques can also be applied to phase-based MRA methods. This application has not been worked out here. Another extension could be the simultaneous acquisition of systolic and diastolic data, for example, to separate arteries and veins. For this purpose, the dummy signal excitations in the method as presented here should be replaced by acquisitions of diastolic data. These extensions require further study. Acknowledgment - The authors wish to acknowledgeP. van Dijk for valuable suggestions and critical comments.
REFERENCES 1. Wedeen, V.J.; Meuli, R.A.; Edelman, R.R.; Geller, S. G.; Frank. L. R.; Brady, Th. J.; Rosen, B. R. Projective imaging of pulsatile flow with magnetic resonance. Science 230:946; 1985. 2. Dixon, W.T.; Du, L.N.; Faul, D.D.; Gado, M.; Rossnick, S. Projection angiograms of blood labeled by adiabatic fast passage. Magn. Res. Med. 3:454; 1986.
3. Nishimura, D.G.; Macovski, A.; Pauly, J.M.; Conolly, SM. Magnetic angiography by selective inversion recovery. Magn. Res. Med. 4:193; 1987. 4. Dumoulin, C.L.; Hart, H.R. Magnetic Resonance Angiography. Radiology 161:717; 1986. 5. Ruggieri, P.; Laub, G.; Deimling, M.; Masaryk, T.; Medic, M. MR angiography of the intracranial vasculature. Abstracts SMRM; 1988: p, 200. 6. Groen, J.P.; De Graaf, R.G.; Van Dijk, P. MR angiography based on inflow. Abstracts SMRM; 1988: p. 906. 7. Keller, P.; Drayer, D.; Fram, E.; Williams, K.; Dumoulin, C.; Souza, S. Initial experience with very thin slice two-dimensional MR angiography. Abstracts SMRM; 1989; p. 103. 8. Dumoulin, C.L.; Souza, S.P.; Walker, M.F.; Wagle, W. Three-dimensional phase contrast angiography. Magn. Res. Med. 9:139; 1989. 9. Bailes, D.R.; Gilderdale, D.J.; Bydder, G.M.; Collins, A.G.; Firmin, D.N. Respiratory ordered phase encoding (ROPE): A method for reducing motion artifacts in MR imaging. .I: Comput. Assist, Tomogr. 9:835; 1985. 10. Cho, M.H.; Kim, W.S.; Cho, Z.H. CSF flow artifact reduction using cardiac style ordered phase-encoding method. Magn. Res. Imaging 8:395; 1990. 11. Lenz, G.W.; Haacke, E.M.; White, R.D. Retrospective cardiac gating: A review of technical aspects and future directions. Magn. Res. Imaging 7:445; 1989.
0730-725x/92 $5.00 + .oo 1992 Pergamon Press plc
1992
copyrigt0
l Original Contribution
MR ANGIOGRAPHY WITH PULSATILE FLOW R.G. DE GRAAFAND J.P. GROEN Philips Medical Systems, P.O. Box 10.000, Best, 5680 DA The Netherlands To achieve acceptable scan times, current multiple thin slice and 3D MR anglograpby (MRA) methods usually are based on continuous data acquisition, without ECG-synchronlxation. Tbe purpose of this work is to study consequences of pulsatlle blood flow for the 2D inilow method. Arterial blood flow and blood slgnai intensity versus cardiac phase were studied by a 2D phase based method with retrospective cardiac synchronlzatlon. Such studies were performed in different parts of the body and with different excitation flip angles. As expected, a clear relation between intensity enhancement and time dependent flow can be demonstrated. The raw data of these multiphase studies was used to simulate alternative htflow MRA data acqulsltion strategies to improve image quality, without the excessive increase in scan thne huplled by standard cardiac trlggerhtg. The alternatives investigated were data collection during part of the cardiac cycle and cardiac-ordered phase encoding. Simulation results indicate that the best results are obtained by a combination of both strategies. This method was implemented on Philips Gyroscan systems to compare it with standard nontrlggered 2D inflow in practical MRA studies. For highly pulsatile flow, much better MR angiograms were obtained in this way. Keywords: MRI; Angiography; Pulse sequences; Puisatile blood flow.
The early MR angiography (MRA) methods took into account the pulsatile nature of arterial blood flow by applying ECG-triggered acquisitions. l4 Recently, the emphasis has shifted to nontriggered MRA methods, which can provide 3D angiographic data in an acceptable scan time. The 2D and 3D inflow methods5-7 are both based on a short TR gradient echo sequence, which saturates proton spins in stationary tissue and highlights fresh unsaturated proton spins flowing into the imaging slice or volume. Generally, velocity-compensated sequences are used to make the phase of the MR signal insensitive to flow velocity. However, during parts of the cardiac cycle with a very low flow velocity, the amplitude of the signal from proton spins in blood will decrease due to saturation of the spins. Still, it has been found possible to obtain MR angiograms, for instance of the carotid and cerebral arteries, by the nontriggered inflow-based methods. Good MRA results have even been shown for the 2D and especially the 3D phase-based subtraction
method’ applied in a nontriggered mode, despite their necessary deviations from true velocity compensation. The best angiographic results by nontriggered methods are obtained for the head and neck. Other regions, for example, the legs, often produce relatively poor angiograms. From the arguments given above, the quality of arterial MR angiograms in different parts of the body is expected to be related to the pulsatility of arterial blood flow. The aim of this work is to further explore this influence, with emphasis on the 2D inflow method. First, our method to study pulsatile blood flow versus cardiac phase will be outlined. Results obtained in different parts of the body will be discussed. On the basis of this information, simulations of the standard nontriggered MRA methods and of modified acquisition schemes will be presented. Our first approach to improve MRA data is a gating strategy, in which data is acquired only during part of the cardiac cycle. During this gate, signals are measured for a number of phase encodings. To preserve a steady state, spin excitations without data acquisition are per-
RECEIVED 519191; ACCEPTED 817191. Address all correspondence to R.G. de Graaf, Philips Med-
ical Systems, P.O. Box 10.000, Best, 5680 DA The Netherlands.
INTRODUCTION
25
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
26
cardiac
cycle
A
B Fig. 1. Magnitude images of a slice through the abdominal aorta. TR = 30 msec, flip angle fW, slice thickness 4 mm. (A) Systolic phase. (B) Diastolic phase.
formed during the remainder of the cardiac cycle. This gating approach has a scan time penalty, determined by the gate width compared to the cardiac period. The second approach is an ordering strategy for phase encodings in relation to the cardiac cycle. The original linear phase-encoding order changes in this approach into interleaved linear k,-sweeps. This method is analogous to respiratory ordered phase encoding (ROPE)9 for respiratory artifact reduction. Very recently, Cho et al. published results on a similar cardiac phase encoding ordering approach for CSF flow artifact reduction. lo Finally, a combination of gating and ordering has been studied. Simulation of the various methods, using the raw data from our pulsatile flow studies, allows us to evaluate the proposed strategies without considering all practical implementation details. The combination of the gating and ordering approaches was implemented on an MR system for investigation of its expected qualities in practice.
caraiac cycle
B Fig. 2. Variation of the MR signal of blood in the internal carotid arteries through the cardiac cycle for excitation flip angles of 30 and 60”. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
PULSATILE FLOW INVESTIGATION A gradient-echo sequence with retrospective gating” has been applied to obtain angiographic data on a single slice for a large number of points in the cardiac cycle. Unlike cardiac triggering, this technique provides steady-state acquisition with the ability to reconstruct
MR angiography with pulsatile flow 0 R.G.
I
I
I
cardiac
0.0
1.0
J.P.
,
-101
I
cycle
DE GRAAF AND
0.0
A
27
GROEN
,
,
cardiac
cycle
/
j 1.0
A I --7---T-----j
I
I
~
I
I
_~- _.
30 deg
_’ . .
.
‘r’?
G ;cardiac 0.0
&A
_
1
1.0
wardiac 0.0
c&FL
1.0
B
Fig. 3. Variation of the MR signal of blood in the abdominal aorta through the cardiac cycle for excitation flip angles of 30 and 50”. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
Fig. 4. Variation of the MR signal of blood in the femoral arteries through the cardiac cycle for excitation flip angles of 30 and 60“. TR = 30 msec, slice thickness 4 mm. (A) Flow curves derived from phase difference data. (B) Magnitude curves.
images for cardiac phases over the complete heartcycle. Thin slice images were acquired with different flip angles in several parts of the human body, such as neck, breast, abdomen, and extremities. The saturation effects of blood and static tissue were studied as a function of flip angle and cardiac phase. Interleaved with phase-en-
coding gradient values, two values of a flow sensitizing gradient were applied, and phase-difference images were calculated to quantify flow perpendicular to the slice. An example of systolic and diastolic magnitude images in the abdomen is given in Fig. 1. Flow curves over the cardiac cycle, obtained by the retrospectively
28
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992 40
0.0
I
I
I
I
30 deg ’ I cardiac
I
I cycle
I 1.0
A I
I. Fig. 5. arteries 30 and curves curves.
I
I
B
I
.
Variation of the MR signal of blood in the popliteal through the cardiac cycle for excitation flip angles of 60”. TR = 30 msec, slice thickness 4 mm. (A) Flow derived from phase difference data. (B) Magnitude
gated phase-based technique described above are given in Figs. 2(A)-5(A) for the carotids, the abdominal aorta, the femoral, and the popliteal arteries, respectively. The corresponding signal amplitude curves, obtained from the same regions of interest in the magnitude images, are given in Figs. 2(B)-5(B). Both the flow curves and the signal amplitude curves are given for excitation flip angles of 30 and 60”.
I -“Y “__
I
I
phase encoding
I
I
mtmber
Fig. 6. Simulation result of the standard nontriggered 2D inflow method for the femoral arteries. TR = 30 msec, flip angle 60“. slice thickness 4 mm. (A) Magnitude image. Large signal intensity variations versus heart phase give rise to ghost artifacts (arrow). (B) Signal variation as a function of phase encoding number.
DISCUSSION OF PULSATILE FLOW RESULTS As can be seen in Fig. 1, there is a considerable difference in image intensity for the abdominal aorta between systole and diastole. During systole, the high flow rate ensures effective inflow of fresh spins. As a result, a high signal intensity is observed for the abdominal aorta. Less efficient refreshment during diastole leads
MR
c
-KY
I
I
angiography with pulsatile flow 0 R.G.
I
I
Phase encoding
nu&er
+KY
DE GRAAF AND J.P. GROEN
-KY
29
Phase encoding
nuntber
+KY
B
6
Fig. 7. Simulation result of the ECG-gated 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. Prominent ghost artifacts are present (arrow). (B) Signal variation as a function of phase encoding number.
Fig. 8. Simulation result of the ECG ordered phase encoding 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. Vessels tend to be blurred (arrow) due to pulsatile flow, especially when larger excitation flip angles are used. (B) Signal variation as a function of phase encoding number.
to a considerably lower end-diastolic signal intensity of the aorta. The flow curves are more or less identical for both flip angles. This was expected because the flip angle
mainly influences signal amplitude. The correlation between the flow and the signal intensity curves can be explained by inflow refreshment. High flow results in an effective refreshment and thus a high signal intensity. For small flip angles, the increase in signal intensity is relatively smaller, but persists during more excitations after a flow peak. The pulsatility of flow depends on the anatomical region. For the carotid arteries (Fig. 2), the average in-
tensity over a vessel cross-section shows relatively small variations over the cardiac cycle. However, intensity inhomogeneities were observed during systole, which are probably due to turbulent flow. Stronger pulsatility of flow, resulting in stronger signal intensity variations, were found for the abdominal aorta and the femoral and popliteal arteries. Large veins such as the vena cava and the lower part of the jugular veins showed some pulsatility effects as well. Experiments described here were performed on healthy volunteers. Note, however, that vascular diseases may increase the pulsatility, resulting in more flow
30
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
artifacts for the nontriggered 2D inflow method. This may be the case for example with atherosclerotic disease,which decreases the flexibility of the vessel walls. In conclusion, signal intensity variations due to pulsatile blood flow are an important problem for inflow based MR angiography methods. In nontriggered applications, artifacts and loss of vascular contrast are due to occur. Phase-based angiography is less sensitive to saturation effects, because generally smaller flip angles are used compared to the inflow technique. However, the phase information is directly related to the velocity of blood, and therefore sensitive to pulsatile flow as well. For such methods, reverse flow will even reverse the contrast. For both types of methods, other effects related to the cardiac cycle, such as accelerated flow and movement of vessels, will add to the artifact level and the loss of vascular contrast.
SIMULATION OF MEA METHODS Raw data of the retrospectively gated scans described above together with the corresponding ECG timing data enabled us to simulate different acquisition strategies for MR angiography on pulsatile vessels. Simulations were performed for the standard 2D inflow method, and for the gating and ordering approaches described in the introduction. The retrospective ordering algorithm was adapted to provide images representative for the different acquisition strategies. Standard retrospective ordering combines data for one specific cardiac phase into an image. In the simulations of MRA methods, a cardiac timing variable was incremented with the repetition time for each subsequent spin excitation. Data corresponding to this cardiac timing variable was selected from the multiphase raw data set. For a nongated measurement, the timing variable was reset to zero each time the length of the cardiac cycle was exceeded. In simulations of the proposed gating method, the timing variable was reset to the beginning of the gate when it exceeded the end of the gate. To simulate ordering of phase encodings reiative to the cardiac cycle, the cardiac timing variable was incremented with the average heart beat interval divided by the total number of spin excitations. Thus, the range of phase encodings is mapped on the cardiac cycle. For the combination of gating and ordering of phase encodings, the cardiac timing variable was initially set to the start of the gate. For each phase-encoding or averaged signal it was incremented with the width of the gate divided by the total number of measurements, and the corresponding data was selected from the multiphase data set. Thus, a mapping of all phase-encoding values on the duration of the gate was achieved.
-KY
phase encoding mmd3er
+Ky
B Fig. 9. Simulation result of the gated sweep 2D inflow method on the femoral arteries. TR = 30 msec, flip angle 60”, slice thickness 4 mm. (A) Magnitude image. (B) Signal variation as a function of phase encoding number.
SIMULATION RESULTS Results for the standard nontriggered 2D inflow method are shown in Fig. 6. A linear order of phase encodings was assumed. To achieve a high contrast for MRA, a data set acquired with a 60” flip angle was utilized for this simulation. Large signal intensity variations versus heart phase, as demonstrated in the retrospectively gated images, give rise to ghost artifacts (arrow) and a low vessel intensity in the nontriggered 2D inflow simulation. Ghost images of vessels occur due to periodical variations in signal intensity as a function of phase-encoding gradient value, as illustrated schematically in Fig. 6(B). The differences in flow pul-
MR angiography with pulsatile flow 0 R.G.
DE GRAAF AND J.P. GROEN
31
between (A) gated sweep and (B) standard nontriggered 2D inflow acquisition of the aorta bifurcation. Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 4 mm.
Fig. 10. A comparison
satility explain why nontriggered angiography of the carotids and the cerebral arteries gives reasonably good results, while artifacts and loss of signal are found for the lower aorta and the illiac arteries. The alternative acquisition strategies were examined on their ability to increase vascular contrast, reduce artifacts, and maintain sharp vessel edges. The discussions are based on the 2D inflow method, but similar arguments can be made for phase-based MBA methods. Simulations were carried out for different parts of the body to investigate the achievable improvements of alternative methods for different characteristics of pulsatile flow. The ECG gating method achieves a reduction of pulsatile flow effects at the cost of scan time. The width of the gate and its delay with respect to the R-wave were varied. This allowed adjustments of the gate in such a way that refreshment of blood is guaranteed for all acquired phase encoding steps. An example of the results is shown in Fig. 7(A). In this case, the gate was about 50% of the cardiac cycle starting at the rising edge of the systolic flow peak. An illustration of the remaining variations in signal intensity is given in Fig. 7(B). Compared to the nongated approach, improved arterial contrast is obtained. However, prominent ghost (arrow) artifacts are still present. The ghost artifacts are enhanced due to the fact that a fixed gate width results in an exactly periodical variation of signals versus phase-
encoding number. This should be compared to a nongated method, where ghosts are smeared to some extent by variations in heart rate. Unlike gating, an ordering approach does not introduce a time penalty. The acquisition order is changed in such a way that a single period of cardiac motion covers the complete range of phase-encodings. In practice, this will be achieved by multiple interlaced sweeps through the range of phase encoding values, synchronized with the cardiac cycle. A linear relationship between cardiac phase and phase-encoding number results, with an adjustable phase offset between the two. An example of the results is shown in Fig. 8. Ordering of the phase-encoding gradient relative to the cardiac cycle results in simulated images free of artifacts, although vessels tended to be blurred (arrow) for very pulsatile flow, especially when larger excitation flip angles are used. This result can be compared to similar observations of blurring for ROPE applied to respiratory motion. In our case, the lower phase-encoding numbers are acquired during the systolic part of the cardiac cycle, which results in improved contrast. However, as can be seen in Fig. 8(B), the higher phase-encoding numbers occur at the end-diastolic phase, where the slowly flowing blood is saturated. Thus, the expected image resolution for vessels is not fully achieved in the phase encoding direction. Cardiac gating with ordering of phase encodings can
32
Fig. 11. A comparison
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
between (A) gated sweep and (B) standard nontriggered
2D inflow acquisition of the popliteal trifurcation.
Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 3 mm. combine the advantages of both methods mentioned previously. In this case, a full sweep of phase-encoding values is performed for each gate. Signals for intermediate phase-encoding values are measured in subsequent cardiac cycles. The gate delay and width can be adjusted for optimal results, depending on the region of interest in the human body. The applied gate ensures enough refreshment of blood, while the ordering of phase encodings eliminates ghost artifacts caused by the remaining signal strength variations. An example of simulation results is shown in Fig. 9(A). Figure 9(B) illustrates the signal variation versus kYfor this method. This approach turns out to be very effective to obtain angiographic images free of artifacts
and with a good vessel definition and contrast. Practical implementation is facilitated by the fixed width of the gate, that is, in fixed phase-encoding sweeps which can be calculated in advance. This method, which will be referred to as Gated Sweep has been investigated in practice, as will be described in the next section. IMAGING RESULTS The Gated Sweep method was implemented on the Philips Gyroscan T5 and ACS systems (Philips Medical Systems, Best, The Netherlands) for comparison with the standard nontriggered 2D inflow method. Using an ECG signal, the Gated Sweep method acquires data dur-
MR
angiography withpulsatileflow0
R.G. DEGRAAFANDJ.P. GROEN
33
Fig. 12. A gated sweep 2D inflow acquisition of the aortic arch. Coronal maximum intensity projection of transversal acquired slices. TR = 30 msec, flip angle 60”, slice thickness 3 mm. Four overlapping partial projections are shown.
ing a gate in the cardiac cycle, while dummy spin excitations are performed in the remaining time. During each data acquisition gate, a sweep through the full range of phase-encoding values is performed. Intermediate phase-encoding values are applied in subsequent cardiac cycles. Depending on the anatomical region to be imaged, gate width and delaly from the R-wave can be adjusted. Typically, a gate width of 50% of the cardiac cycle is selected. Thus, the scan time is increased by a factor of two compared to standard nongated methods. In comparisons, this was compensated by using twice the number of measurements for the standard 2D inflow method. Results of Gated Sweep and standard 2D inflow are shown in Fig. 10 for the aorta bifurcation and in Fig. 11 for the popliteal trifurcation. The delay of the gate relative to the R-top of the ECG was adjusted according to the delay in onset of flow for the vessel of interest. For the aorta and the politeal trifurcation, delays of 150 msec and 250 msec were used, respectively. The gate width is selected to ensure sufficient signal intensity enhancement by refreshment of blood. For 2D inflow measurements, a width of 50% of the cardiac cycle was typically used. Both parameters are not very critical, and fixed values can probably be used for different patients. In case of aorta bifurcation (Fig. lo), the Gated Sweep method results in an increased intensity of the aorta and eliminatiaon of stripe patterns caused by pulsatile flow. Similarly, the arterial structures just below the knee (Fig. 11) are increased in intensity compared
to standard 2D inflow while veins with a rather constant flow are not affected. For MRA of the carotid arteries, the gate width could be adjusted to almost the full cardiac cycle. In that case, pulsatile flow artifacts are only suppressed by ordering of phase encodings and no significant loss of scan time occurs. For cerebral MRA, no significant improvements were observed. The aortic arch is a difficult area for nontriggered MRA methods, since movements of the arch will lead to artifacts. Figure 12 shows results obtained with the Gated Sweep method. In this case, the gate delay was 100 msec and the width 300 msec. Slices were acquired with a transverse orientation. The results of a partial projection clearly display the left and right common carotid arteries and the subclavian arteries. CONCLUSIONS Our study of flow pulsatility influences shows clear differences between different anatomical regions, with important consequences for MRA acquisition methods. Improved methods were developed for strongly pulsatile vessels. An important reduction in artifact level and improvement in vascular intensity has been demonstrated with a combination of cardiac gating and ordering of phase encodings. The gating approach introduces a time penalty of a factor of two, but is much more effective than signal averaging. Results on the Gated Sweep method presented here are preliminary because
34
Magnetic Resonance Imaging 0 Volume 10, Number 1, 1992
clinical evaluation of this method is still to be performed. Choices for the gate width and gate delay need to be evaluated thoroughly in such a study. Similar techniques can also be applied to phase-based MRA methods. This application has not been worked out here. Another extension could be the simultaneous acquisition of systolic and diastolic data, for example, to separate arteries and veins. For this purpose, the dummy signal excitations in the method as presented here should be replaced by acquisitions of diastolic data. These extensions require further study. Acknowledgment - The authors wish to acknowledgeP. van Dijk for valuable suggestions and critical comments.
REFERENCES 1. Wedeen, V.J.; Meuli, R.A.; Edelman, R.R.; Geller, S. G.; Frank. L. R.; Brady, Th. J.; Rosen, B. R. Projective imaging of pulsatile flow with magnetic resonance. Science 230:946; 1985. 2. Dixon, W.T.; Du, L.N.; Faul, D.D.; Gado, M.; Rossnick, S. Projection angiograms of blood labeled by adiabatic fast passage. Magn. Res. Med. 3:454; 1986.
3. Nishimura, D.G.; Macovski, A.; Pauly, J.M.; Conolly, SM. Magnetic angiography by selective inversion recovery. Magn. Res. Med. 4:193; 1987. 4. Dumoulin, C.L.; Hart, H.R. Magnetic Resonance Angiography. Radiology 161:717; 1986. 5. Ruggieri, P.; Laub, G.; Deimling, M.; Masaryk, T.; Medic, M. MR angiography of the intracranial vasculature. Abstracts SMRM; 1988: p, 200. 6. Groen, J.P.; De Graaf, R.G.; Van Dijk, P. MR angiography based on inflow. Abstracts SMRM; 1988: p. 906. 7. Keller, P.; Drayer, D.; Fram, E.; Williams, K.; Dumoulin, C.; Souza, S. Initial experience with very thin slice two-dimensional MR angiography. Abstracts SMRM; 1989; p. 103. 8. Dumoulin, C.L.; Souza, S.P.; Walker, M.F.; Wagle, W. Three-dimensional phase contrast angiography. Magn. Res. Med. 9:139; 1989. 9. Bailes, D.R.; Gilderdale, D.J.; Bydder, G.M.; Collins, A.G.; Firmin, D.N. Respiratory ordered phase encoding (ROPE): A method for reducing motion artifacts in MR imaging. .I: Comput. Assist, Tomogr. 9:835; 1985. 10. Cho, M.H.; Kim, W.S.; Cho, Z.H. CSF flow artifact reduction using cardiac style ordered phase-encoding method. Magn. Res. Imaging 8:395; 1990. 11. Lenz, G.W.; Haacke, E.M.; White, R.D. Retrospective cardiac gating: A review of technical aspects and future directions. Magn. Res. Imaging 7:445; 1989.
