447

Review Digital Subtraction

Angiography: Principles and
Pitfalls of Image Improvement
Techniques

David C. Levin1 The technology of imaging methods in digital subtraction angiography (DSA) is
RobertM. Schapiro2 discussed in detail. Areas covered include function of the video camera in both interlaced
Lawrence M. Boxt’ and sequential scan modes, digitization by the analogto-digftal converter, loganthmic
signal processing, dose rates, and acquisition of images using frame integration and
American Journal of Roentgenology 1984.143:447-454.

Lance Dunham1
pulsed-sequential techniques. Also discussed are vanous methods of improving Image
Donald P. Hamngton1
content and qualfty by both hardware and software modifications. These include the
David L. Ergun1
development of larger image intensifiers, larger matrices, video camera Improvements,
reregistratIon, hybrid subtraction, matched fiftering, recursive filtering, DSA tomography,
and edge enhancement.

Since the original digital subtraction angiography (DSA) systems were introduced
in the late 1970s, some important advances in imaging technology have occurred.
As a result, a variety of methods of improving DSA quality have been either
proposed or implemented. This has created some confusion among radiologists
who must make decisions regarding purchase and use of such systems, but who
generally lack the physics or engineering background needed to properly evaluate
them. Perhaps as a result, one can note that many recent publications on DSA fail
to describe the technical factors or equipment specifications used in performing
procedures, even though the very purpose of these studies is to assess the efficacy
of DSA as an imaging method.
First, we will review some of the basic principles applicable to all television (TV)
and DSA imaging techniques; second, we will discuss the various methods of
improving image quality (some already established, others under development) and
point out some of their associated pitfalls. These image improvement methods
include frame integration, sequential video scanning, video camera modifications,
the use of larger image intensifiers and matrices, reregistration, hybrid subtraction,
recursive filtration, matched filtration, DSA tomography, and edge enhancement.
The discussion will primarily emphasize acquisition and processing of serial-mode
images at 1-10 frames/sec, the type of imaging generally used by radiologists for
noncardiac applications. Image storage technology will not be addressed here, nor
will cardiac imaging, which is often performed at real-time frame rates (30 frames/
sec) and presents a somewhat different set of problems.
Received February 14, 1984; accepted after re-
vision May 8, 1984.
Basic Principles
This work was supported in part by U.S. Public
Health Service grants GM1 8674 and HL20895. The front-end components for video fluoroscopy include the x-ray tube, table,
1Department of Radiology, Harvard Medical
image intensifier, distributor, ins, TV camera, and suitable monitor for display. The
School, 25 Shattuck St., Boston, MA 02115. Ad-
dress reprint requests to D. C. Levin.
x-ray pattem impinges on the input phosphor of the image intensifier and is
2Pheps Medical Systems, Shelton, CT 06484. converted to a light pattem at the output phosphor. This light image is conveyed
to the video camera by tandem lenses or fiberoptic coupling. The front of the
AJR 143:447-454, S.pt.mber 1984
0361 -803X/84/1433-0447 camera contains the target, a thin layer of tiny lead monoxide elements, which
American Roentgen Ray Socsety become positively charged in proportion to the amount of light they receive from
448 LEVIN ET AL. AJR:143, September1984

1000
3-BAR OBJECT ON
IMAGE INTENSIFIER
100
SIGNAL-TO-
NOISE RATIO
10

J1J1J1 IDEAL
0 10 20
VIDEO
SIGNAL BAND WIDTH (MHz)
LOW BANDWIDTH
VOLTAGE
Fig. 2.-Mdeo camera signal-to-noise ratio as function of bandwidth. This
curve was plotted from tests of cameras prior to 1980. cameras of more recent
design would show higher and somewhat steeper curve. (Reprinted from [3].)
HIGH BANDWIDTH

Fig. 1 -Voltage variation as video camera electron beam sweeps along

single raster line and encounters 3-bar object. Top tracing shows ideal voltage
pattern to create TV image of this object. Middle tracing shows voltage pattern encounters varying degrees of brightness or darkness repre-
that might occur if camera had extremely low bandwidth. There would be no sented by the charge pattern on the target. This voltage
video noise but very poor resolution. Lower tracing shows pattern that might
occur if camera had extremely high bandwidth. Edges of object would be
variation is of course caused by both information-carrying
American Journal of Roentgenology 1984.143:447-454.

detected very sharply, but because of high system sensitivity, image would be signal(e.g., iodine in an artery) and noise (particularly quantum
very noisy. and video noise)-the former desirable, the latter undesirable.
The bandwidth of the video camera is defined as its ability to
vary its voltage up or down rapidly in response to interfaces
between opaque or less opaque structures. Bandwidth is
the image intensifier [1 2J. A thin electron beam sweeps
,
usually expressed in megahertz (1 MHz = 1 ,000,000 cycles/
across the target horizontally, providing a current flow path sec). To image very small vessels, a high bandwidth is ob-
for these charged elements. This creates a flow of current viously necessary. However, if the bandwidth is too high, the
with a voltage signal proportional to the amount of discharge camera becomes so sensitive that every bit of quantum or
that occurs at each successive point. Most video cameras video noise is also detected, with resultant degradation of the
create images by scanning the target in a “raster” pattem of image. This compromise is illustrated in figure 1 SNR de- .

525 horizontal linear sweeps 30 times each second (thus, creases as video bandwidth increases (fig. 2).
about 33 msec is required for each complete TV image). At The video signal is then passed through an analog-to-digital
the end of each horizontal linear sweep, the electron beam converter (ADC). As the electron beam sweeps across each
retraces back to the opposite side of the target to begin horizontal raster line, the ADC samples the signal at specified
scanning the next line. intervals and assigns a binary number according to the analog
To perceive continuous motion without flicker, the human voltage level at each sampling point. ADCs currently used in
eye must integrate at least 48 images/sec [2]. If only the 30 DSA systems should have at least an 8-bit depth or preferably
video images/sec were created, a distracting flicker would a 10-bit depth (meaning that at each sample point, it can
appear on the TV screen during fluoroscopy. To prevent this, assign any one of256 or 1 024 values, respectively, depending
an “interlaced” pattern is used. In interlaced scanning, the on the voltage level). The number of samples along each
electron beam completes its sweep of line 1, but then instead raster line determines the number of horizontal pixels in the
of retracing back to begin at line 2, it skips to line 3, then line image matrix. Horizontal resolution of a DSA system is thus
5, line 7, and all 262 odd lines. This 262-odd-line field is determined to a large extent by the sampling frequency of
completed in /eo sec. The beam then retraces back to the top the ADC and the video bandwidth. On the other hand, vertical
of the target to commence scanning line 2, then line 4, and resolution is determined by the number of horizontal raster
all 262 even lines. This also takes l/6o sec, so that a completed lines in the video scan. Most medical-grade TV systems
TV frame consisting of odd and even fields is created each operate at a 5 MHz bandwidth with 525 scan lines, while the
1/3 sec. Because 60 fields are presented to the angiogra- ADCs sample 512 times across the active part of each line.
pher’s eye each second (even though they represent only 30 Each pixel corresponds to a specific site in the image proces-
interlaced frames), no flicker is apparent. Each of the 60 TV sor memory. Once digitization of the analog TV signal has
fields/sec contains continuously upgraded information re- been accomplished, this digital information is transferred to
ceived from the output phosphor of the intensifier. Early DSA the corresponding position in the memory so that subsequent
units used cameras designed for fluoroscopic applications, subtraction, contrast enhancement, or other types of image
with interlaced readout only and signal-to-noise ratios (SNR5) processing can be carried out.
on the order of 200-400:1. Because of the exponential attenuation of x-rays in the
During each horizontal sweep of the video camera electron patient’s body, most DSA systems provide some form of
beam, a rapidly varying voltage signal is created as the beam logarithmic processing of the video signal to compensate for
AJR:143, September 1984 TECHNIQUES IN DSA 449

varying tissue thickness [1 4]. A logarithmic

, amplifier accom- operation is a desirable feature in DSA units being considered
plishes this by amplifying signals in inverse proportion to the for purchase. It should be noted that frame integration re-
log of their intensity; the net effect is that images of contrast- quires an increase in bit depth of the image processor mem-
filled arteries will be uniform despite variation in overlying ory. For example, if the memory is 8 bits deep, it can process
tissue thickness. Logarithmic processing can be carried out only a single frame having 256 different shades of brightness.
either before or after digitization, depending on specific sys- To integrate a second frame requires 256 additional binary
tern design. values (i.e., a 9-bit memory). Four-frame integration requires
It is important to bear in mind that for intravenous DSA a 10-bit memory; eight frames require 11 bits of memory
imaging, we deal with very low intensity iodine signals due to depth; and so on.
dilution of the contrast material. The creation of adequate
diagnostic images thus requires noise reduction to the lowest
possible level so these weak iodine signals will not be inter- Sequential Video Scanning with Pulsed Radiation
fered with. One important way to reduce noise is to use higher
A major variation of the older, interlaced method of readout
x-ray photon flux than in standard fluoroscopy. This requires
of the video camera target has already been introduced by
delivery of a higher dose of radiation to the face of the image
most manufacturers [1 3-1 5]-the use of sequential (some-
intensifier. The creation of a single frame on a TV monitor
times termed “progressive”) readout after a short pulsed
each 1/3 sec during routine fluoroscopy requires about 2-4
radiation exposure (hereafter referred to as the pulsed-se-
R (0.5-1 .0 nC/kg). The creation of a single serial mode DSA
quential method). With this technique of creating TV images,
image requires about 1 000 MR or 1 mA (0.26 SC/kg).
a short but intense pulse of radiation (generally 1 0-1 00 msec)
Whereas routine fluoroscopy is generally performed at 1-3
creates the image on the image intensifier, which is then
mA, serial-mode DSA images are usually obtained at 200-
American Journal of Roentgenology 1984.143:447-454.

imprinted on the video camera target. The generator then

1 300 mA [5-7]. With intraarterial injection, there is less dilu-
shuts off, but the image remains stored on the target until it
tion of contrast material. This allows use of a lower radiation
is discharged by the scanning electron beam. Scanning of the
dose while providing superior DSA image quality.
target commences after generator shutoff. Instead of the
The above basic principles apply to DSA systems in current
interlaced method of scanning first the odd lines in /o sec
use. We will now consider methods of improving image quality
and then the even lines in another /o sec, the sequential
that involve either modification or extension of the capability
method scans all 525 lines in consecutive order in /o sec to
of the basic imaging system.
create an entire TV frame. Radiation doses with multiple-
frame integration and the pulsed-sequential technique are
comparable.
Frame Integration
There are several advantages of pulsed-sequential imaging,
Because of the poor SNRs of early video cameras, it was the most important of which is significant shortening of the
found that high-quality DSA images could not be obtained exposure time. This helps alleviate the problem of motion and
using single TV frames. Too much noise was present. It was vessel pulsation that occurs when frame integration is used.
instead necessary to use integration, or averaging, of multiple Because frames are not wasted waiting for the video signal
frames (typically two to eight) to improve the effective SNR to stabilize, optimum dose efficiency is achieved [1 6]. If
[5, 8-1 1]. If n frames are integrated, SNR improves by a installation of a biplane DSA laboratory is contemplated, a
factor equal to the square root of n [1 2J. Frame integration pulsed-sequential system allows the use of alternate pulsing
can thus be considered a method of weighting the incoming of the two tubes to minimize scatter [1 7]. Integration of
information to enhance the desirable signals at the expense individual pulsed images can be used with the pulsed-sequen-
of the undesirable ones. This technique allows the use of tial method [1 3, 1 8] and may prove to be of benefit in
lower powered generators, since exposure times are gener- enhancing image quality. Not all systems currently provide
ally long (1 50-500 msec). this capability, however.
The use of integration techniques is not necessarily limited Use of the pulsed-sequential method alters system require-
to averaging some number of TV frames to create a single ments significantly. Without integration, the images tend to
DSA image. It is also possible to integrate an extended series be noisier. Therefore, a higher SNR camera is needed to
of discrete sequentially scanned images to try to improve reduce video noise. The shorter exposure times necessitate
SNR. Sequential scanning will be described below. the use of significantly higher amperage levels [1 3, 1 4J. This,
A major drawback of frame or image integration is that it in turn, requires higher power generators and x-ray tubes.
lengthens the effective exposure time. As a result, motion Equipment manufacturers often encourage radiologists to
artifacts and arterial pulsations are likely to degrade image acquire DSA capability by the simple expedient of adding the
quality, especially in studying actively pulsating high-flow ves- necessary hardware on to older existing radiographic and
sels, such as the thoracic aorta and pulmonary arteries. fluoroscopic rooms. This can be done if one is willing to use
Another disadvantage is that after exposure begins, the first only frame integration as the primary method of reducing
few TV fields are wasted to allow the video signal level to noise. However, if one prefers to use shorter exposure times
stabilize, thus adding to patient exposure and x-ray tube load. with pulsed-sequential imaging technique, high-SNR video
Despite its drawbacks, frame integration remains a basic cameras and higher powered x-ray generators and tubes will
method of image improvement; the capacity to perform this be needed.
450 LEVIN ET AL. AJR:143, September1984

There is at present some disagreement among manufac- significantly reduced. Camera SNR is determined primarily by
turers as to which basic imaging method is most desirable. the ratio of signal current to preamplifier noise. Since the slow
Some rely on interlaced scanning with frame integration to scan technique reduces signal current, proper camera design
provide sufficient SNR for good image quality. Others use incorporating high inherent signal current and low preamplifier
single-frame imaging with pulsed-sequential technique and do noise becomes vital to maintain acceptable SNR.
not offer the option of frame integration. Still others provide Another drawback of the slow scan method is that no more
the capability for both pulsed-sequential imaging and integra- than 2 or 3 images/sec can be obtained. An additional problem
tion of discrete frames. It is not yet clear whether either of created by the 1 024 x 1 024 matrix is that of transferring
these methods is consistently better than the other, or information to disks or tape for storage. This process is
whether one might be preferable for some vascular beds already quite slow when digital storage methods are used,
whereas the second is preferable in other vascular beds. and a fourfold increase in the amount of information to be
Further research in this area is needed. transferred will make matters worse.
Because of the problems detailed above, it is not yet clear
whether imaging with larger intensifiers and 1 024 x 1024
Larger Image Intensifiers and Matrix Sizes pixel matrices will be as readily achievable as one is led to
believe by reading the promotional literature.
A drawback of early DSA systems was that they were
limited to the 6 or 9 inch (1 5 or 22.5 cm) field size of most
image intensifiers. This is often insufficient coverage for ab-
Video Camera Improvements
dominal or lower extremity studies. Several manufacturers
have therefore developed large (and expensive) 1 4 inch (35 As has been pointed out by Riederer [22], it is difficult to
American Journal of Roentgenology 1984.143:447-454.

cm) or 1 6 inch (40 cm) intensifiers to allow full angiographic obtain high-quality DSA imaging with older video cameras
field coverage [1 9, 20]. If matrix size remains constant (e.g., operating with an SNR in the 1 00:1 or 200:1 range. Newer
51 2 x 51 2 pixels) but anatomic field coverage increases from cameras with SNRs well in excess of 1 000:1 at 5 MHz
9 inches (22.5 cm) to 14 inches (35 cm), spatial resolution bandwidth have recently become commercially available.
must decrease. Price and James [1 1 estimated that a 51 2 x Some of these, and most future cameras, will use the Am-
51 2 pixel matrix can resolve about 1 .1 -1 .2 line pairs (LP)/mm perex XQ4502 frog’s head plumbicon [4, 23, 24]. Like older
on a 9 inch (22.5 cm) field, but only 0.7-0.8 LP/mm on a 14 versions of the plumbicon, this unit has a diode electron gun,
inch (35 cm) field. which improves resolution and dynamic range (the ability to
To try and maintain satisfactory spatial resolution with a respond to wide variations in image brightness without be-
large-intensifier-based system, some equipment manufactur- coming saturated) and reduces lag (persistence of some
ers are now developing 1024 x 1 024 pixel matrices. As matrix image information on the target after discharge by the electron
size increases, the area included in each pixel decreases and beam sweep). The target of the XQ4502 plumbicon has a 26
resolution improves. To achieve this number of pixels in the mm usable diameter, which gives it an area about 50% larger
vertical dimension, the camera scan pattern must have 1049 than that of the older tubes. This larger target scan area
raster lines instead of 525. The larger matrix might seem to improves resolution. It also increases signal current, because
be a simple way to improve resolution, but this approach has the latter is directly proportional to scan area. The peak signal
certain drawbacks that are not generally appreciated. Since current is 3500 nA-almost double that of the most advanced
there are four times as many pixels to be imaged, a substantial tubes previously used. The target photoconductive layer is
increase in video bandwidth and frequency of analog-to-digital thinner, thereby decreasing scatter and lateral light leakage.
conversions would be necessary if the standard scan readout When this tube is incorporated into video cameras having
time of 33 msec were maintained. However, as indicated improved preamplifier design that limits noise to less than 1
earlier, an increase in video bandwidth results in lower SNR, nA, a very high SNR can be achieved. This is particularly
which tends to defeat the whole purpose of the 1 024 x 1024 useful when 1 024 x 1 024 pixel matrices are used in DSA
matrix (improved resolution). One solution to this problem is imaging since, as indicated earlier, the slow video scan tech-
the use of a slow video scan technique described by Ergun nique needed for the larger matrix might otherwise reduce
et al. [21]. After the short pulse of radiation stores an image SNR to unacceptable levels.
on the video camera tube target, sequential scanning by the It should be apparent that these advances in camera design
electron beam commences at a reduced speed. Instead of are of great benefit in attempting to maximize DSA image
the usual 33 msec readout, it is increased fourfold to about quality, particularly when using pulsed-sequential systems
133 msec. With this slower sweep speed of the beam, the and 1024 x 1024 pixel matrices. Radiologists considering
frequency with which the camera must rapidly vary its voltage purchase of such systems are well advised to include a state-
up or down need not increase, nor would it be necessary to of-the-art video camera in their planning, despite its consid-
increase the frequency of analog-to-digital conversions. A erably higher cost. In the near future, we can expect signifi-
lower bandwidth can be used and SNR should be maintained. cant improvements in ADCs and camera preamplifier design,
However, the maximum signal current obtainable from the such that 1000:1 SNRs will be achievable at wider bandwidths
video camera is inversely proportional to the scan time. If and conventional scan times of l/3 sec. This will increase the
scan time is increased fourfold, average signal current will be feasibility of imaging with 1024 x 1024 matrices.
AJR:143, September1984 TECHNIQUES IN DSA 451

Reregistration (Pixel Shifting) I 000

The most common problem in clinical DSA imaging is failure

to obtain exact superimposition of all anatomic structures on
the mask and contrast images. The resulting misregistration I 00
artifacts are caused by swallowing, movement of bowel gas,
or other types of voluntary and involuntary patient motion MASS
during the few seconds between the mask and contrast
images. To try to alleviate this problem, all DSA systems allow ATTENUATION 10
remasking-the selection of an alternate mask image, usually COEFFICIENT
somewhat later in the filming sequence and temporally closer
(cm2/g)
to the contrast image. Artifacts caused by slight motion or
inherent instability of raster-line positioning can also be re- I .0
duced by reregistration, or pixel shifting. This is a software
modification that allows shifting of pixel information horizon-
tally, vertically, or obliquely to improve superimposition of 0. I
mask and contrast images. Assume, for example, that pixel
A has a DSA number (determined by the binary number 0 40 80 120
originally assigned by the ADC to the site) of 255 and that
pixel B just to the right has a DSA number of 1 55. If the PHOTON ENERGY (keV)
computer is instructed to shift the image one full pixel to the Fig. 3.-Variation in mass attenuation coefficients of iodine, bone. and
American Journal of Roentgenology 1984.143:447-454.

right, pixel B’s new DSA number becomes 255, while pixel A muscle with changes in photon energy of x-ray beam. As photon energy
assumes whatever number had been present in the pixel to increases, absorption ofx-rays by these three SUbstanceSdecreases. However,
degreeofdecrease differs among them. Thus age obtained at low kovoftage
its immediate left. This shift is of course performed throughout will show same structures as one obtained at hIgh kilovoltage, but proportional
the entire matrix. If the computer is instead instructed to shift changes in video density from low to high kilovoltage w8 differ for the three
the image 0.1 pixel to the right, it calculates the difference substances. (Reprinted from [3).)
between DSA numbers in A and B (1 00 units), multiplies by
0.1 and adds it to B. Thus, pixel B’s new DSA number
,

becomes 165.
Clinical experience with this technique has shown that it is subtraction, the 1 30 kV image is first weighted (in this case
simple to use and may be of value in salvaging some studies multiplied) by a factor of about 1 .33, the subtraction process
that otherwise would have been nondiagnostic [25]. One must will cause the soft-tissue signals to cancel each other out as
recognize, however, that any type of image manipulation may well, leaving only bone and iodine images. The use of this
create artifacts. type of image processing may make it possible to eliminate
gas and soft-tissue images, thus reducing misregistration
artifacts due to swallowing or bowel gas motion. Bone signals
are then eliminated by standard temporal subtraction (sub-
Hybrid Subtraction
traction of a mask from a contrast-filled image), leaving only
Hybrid subtraction is a combination of standard temporal the desired iodine signal.
and dual-energy subtraction [26, 27] that has recently been Van Lysel et al. [28] and Foley et al. [29] have used hybrid
attempted clinically as an adjunct in DSA imaging [28, 29]. subtraction in preliminary clinical DSA studies and have noted
Dual-energy subtraction is based on the principle that iodine, certain drawbacks inherent in this technique. First, the iodine
bone, and soft tissue each attenuate x-rays to different de- signal intensity is somewhat reduced by the energy subtrac-
grees at high and low photon energies (fig. 3). For example, tion process. Second, since four separate images contribute
a typical DSA frame obtained at 70 kV during the contrast- to each hybrid subtraction (as opposed to two for standard
filled phase contains bone, soft tissue, gas, and weak iodine temporal subtraction), the hybrid image will contain more
signals. If another frame is then obtained less than 50 msec noise. Thus the overall SNR of the process is reduced. It has
later at 130 kV, this frame will show virtually the same been estimated that the SNR of hybrid subtraction is only
anatomy but with about an 80% reduction in iodine signal, a 35%-40% that of temporal subtraction [28, 30]. To improve
40% reduction in bone signal, and a 25% reduction in soft- SNR of their hybrid images, both Van Lysel et al. [28] and
tissue signal, compared with the earlier 70 kV frame. Gas will Foley et al. [29] had to utilize 4-frame integration of both
attenuate very little of the x-ray beam at either energy, and mask and contrast hybrid images. While this improves SNR,
the gas signal (almost 0 attenuation) will be virtually identical it lengthens effective image acquisition time and could thus
in both images. If the information stored in the image proces- introduce arterial or patient motion artifacts. Theoretically,
sor memory from the 130 kV frame is subtracted from that of these artifacts should not appear on these images because
the 70 kV frame, the gas images will effectively cancel each they are cancelled by the hybrid subtraction process. How-
other out, a small soft-tissue signal will remain, and significant ever, Foley et al. [29] did in fact note the presence of unex-
bone and iodine signals will also remain. If, instead of simple plained soft-tissue motion artifacts in some of their studies.
452 LEVIN ET AL. AJR:143, September1984

Another obvious drawback is the doubling of the number of

exposures required to obtain pairs of images. Patient dose
and tube loading are thereby increased. Finally, the complexity
of the DSA equipment is increased. The image processor
must have increased flexibility and storage capacity [28].
Generator switching, changing of x-ray beam filters, and
stabilization of video signal levels must all be accomplished DSA
within extremely short time intervals. Van Lysel et al. [28]
VALUES
used a constant potential generator (Philips Optimus M200)
for their study. Many noncardiac angiographic laboratories,
TIME (sec.)
however, are not equipped with constant potential genera-
tors.
Foley et al. [29] estimated that hybrid subtraction produced
diagnostic improvement in about 20% of their small group of
clinical studies. Van Lysel et al. [28] stated that their results SOFT TISSUE ARTERIAL
PIXEL PIXEL
were inconclusive. The future potential of this technique is
therefore unclear. Whereas it can improve image quality in
some cases, we do not yet know how costly or reliable it will Fig. 4.-Matched filtering. Upper solid curve in diagram a is plot of contrast
be, or whether the improvements might be achieved by other density over time in region-of-interest placed over major artery shortly after
intravenous contrast injection. Lower dashed curve is created by subtracting
more readily available methods. constant from all points along upper curve. Constant is such that DSA numbers
before and during passage of contrast bolus will have average value of zero
American Journal of Roentgenology 1984.143:447-454.

(i.e., shaded regions above and below horizontal line are equal in area). This
Matched Filtering constant is then subtracted from DSA values in each pixel throughout matrix
as imaging run proceeds. In hypothetical pixel in soft-tissue area where no
Matched filtering, recently described by Riederer et al. [31, arteries are present (represented by diagram b), DSA values primarily reflect
random noise. By subtracting constant, matched curve well below horizontal
32], is a postprocessing method of temporal integration of a line is created. In another hypothetical pixel located in center of artery (repre-
sequence of subtracted images that have all been weighted sented by diagram C), DSA values initially reflect random noise but then rise to
peak as contrast bolus passes. By subtracting constant, lower matched curve
to emphasize the iodine signals and suppress background
5 created, but this curve’s peak rises above horizontal line. If, in creation of
structures and noise. After obtaining a 1 0- to 1 5-sec run of final image, only those DSA values above level of horizontal line are used,
subtracted images, matched filtering was then performed off- image will contain only iodine signals with noise filtered out.
line by choosing a region-of-interest within a major artery and
plotting a curve of contrast density as a function of time. The
resulting data points were fit by the least-squares method to
a smooth bell-shaped curve (gamma variate). A constant value studies [31 32], although
, it did improve images in individual
was subtracted from all points along the curve; this constant cases. It might prove to be particularly appropriate for study-
was chosen so that the resulting second curve would have a ing extremity arteries where motion of gas-containing struc-
mean value of 0 (fig. 4). Next, the same constant was sub- tures (e.g., larynx and bowel) or the arteries themselves is
tracted from each image in the sequence. The weighted not a problem. This temporal filtration approach to image
subtracted images in the run were then integrated to form a improvement seems promising, but further investigation is
single final matched filtered image. By subtracting the con- needed.
stant from each already-subtracted image in the run, much of
the remaining noise and background structures would be Recursive Filtering
removed. The stronger iodine signals would also be some-
what suppressed in individual images, but the final integration Recursive filtering is another temporal filtering technique
of all signal information would result in an enhanced iodine that can be performed with an older, low-SNR video camera,
signal on the final filtered image. using continuous fluoroscopy at 3-30 mA to obtain 30 inter-
Riederer et al. [32] estimate that matched filtering produces laced TV frames/sec [33-36]. Integration of a short and long
a 50% reduction in noise. Although their original studies were sequence of frames from the same run are carried out by a
processed off-line, they indicate that an inexpensive modifi- dual-memory digital recursive filter. For example, in one such
cation of video processor hardware could allow it to be done scheme [35], the short-sequence filter integrates frames dur-
on-line. Interestingly enough, they found that simple integra- ing a 2-sec period corresponding to peak opacification of the
tion of the sequence of unweighted subtracted images pro- arteries. The long-sequence filter integrates frames during the
duced almost the same degree of noise reduction as matched same 2 sec plus the preceding 6 sec. The latter thus acquires
filtering. much image information before the arrival of the contrast
As might be expected with a technique using extensive bolus. The signals in the two filters are weighted so that their
temporal integration of images acquired during a 10- to 15- sum equals 0. Both filters thus contain background struc-
sec DSA run, motion artifacts can be a significant drawback tures, noise, and artifacts caused by certain types of repetitive
(although some of the minor ones can be suppressed by the motion, such as arterial pulsations. However, the short-se-
filter). No consistent improvement in image quality could be quence filter contains strong iodine signals, whereas the long-
demonstrated by matched filtering in the preliminary clinical sequence filter (which has been more heavily weighted with
AJR:143, September1984 TECHNIQUES IN DSA 453

how effective these would be. Another disadvantage is the

extensive equipment modifications that would be required.
Many DSA studies are now being done in laboratories that
are also used for standard catheter angiography. Virtually
none of these laboratories have tomographic capability, and
it is doubtful that they could be satisfactorily converted. The
cost of designing and building entirely new angiographic
mounting units to incorporate tomography might well be
____ I DSA Value
prohibitive. Thus, DSA tomography must at present be con-
sidered purely an experimental technique.

Edge Enhancement

Edge enhancement is another software-controlled method

of image processing that sharpens borders of contrast-filled
vessels. Images are mathematically processed as a “kernel”
Instantaneous (a square box 3-9 pixels on a side) sweeps across the image
rate of change and detects decreases in density as it crosses the borders of
vessels. Calculation of the first derivatives of the curve of
DSA numbers along this sweep line produces large defiec-
tions at the borders of vessels, corresponding to a rapid rate
American Journal of Roentgenology 1984.143:447-454.

of change in density at these borders (fig. 5). The computer

is instructed to enhance those pixels where the rate of change
Fig. 5-Edge enhancement. To enhance unsharp blood vessel borders,
is greatest. In effect, this creates a line along the vessel
square kernel” sweeps across image. As it crosses one border of artery, there
is abrupt change in DSA value. Another abrupt change in reverse direction border, giving it a discrete rather than diffuse edge. This
occurs as kernel crosses opposite border. Instantaneous rate of change in sharpens the border and makes it more amenable to quanti-
DSA values is greatest in those pixels along edges of vessels. By detecting
these peaks and enhancing signals of pixels where peaks occur, vascular
fication of vessel diameter and stenosis. However, edge
borders can be made to appear sharper. enhancement introduces competing artifacts that may lead to
a generally noisier image. Also, there is the possibility that
diagnostic information, such as wall irregularity due to athero-
sclerosis, might be masked by the sharpening process. This
the earlier non-contrast-containing images) contains weaker technique seems promising, at least as an adjunct to standard
iodine signals. Subtraction should thus yield an iodine-only imaging techniques.
image. This review has focused on the technology of producing
An advantage of recursive filtering is that it can be used and improving images in DSA. Although this technology is
with a relatively basic image intensifier-TV combination, now quite sophisticated, a number of questions pertaining to
such as might be found in older angiographic or fluoroscopic equipment still remain to be answered. Is pulsed-sequential
laboratories. As is true with matched filtering, however, re- imaging technique preferable to frame integration, or vice
cursive filtering is susceptible to patient motion artifacts. The versa, or should both capabilities be available for use as
technique has not yet been shown to substantially improve circumstances dictate? To what degree will advances in video
on conventional DSA imaging [35].
camera and ADC design improve image quality? Are 14-16
inch (35-40 cm) image intensifiers desirable or cost-effective?
DSA Tomography Will higher resolution DSA imaging with 1 024 x 1 024 pixel
matrices prove feasible? Can less sophisticated DSA systems
DSA tomography has been studied experimentally by coupled to older fluoroscopic units produce acceptable im-
Kruger et al. [37]. They modified a multidirectional tomo- ages using recursive filtering and other processing tech-
graphic unit by placing an image intensifier with a video niques? What will be the role of hybrid subtraction? Will
camera beneath the Bucky assembly. A circular tomographic postprocessing methods like reregistration, edge enhance-
motion at 1 revolution/i .5 sec was used. Because of the ment, and matched filtering significantly enhance image qual-
relatively long exposure time required to allow completion of ity? The answers to these questions await further research
the rotary motion, recursive filtration was used. Their experi- and development in which both radiologists and commercial
mental images, obtained in living dogs, indicated that the manufacturers should participate.
technique is capable of isolating contrast-filled arteries within
a single anatomic plane, thereby alleviating the common
problem of vessel overlap, while motion artifacts and back- ACKNOWLEDGMENTS
ground structures were largely eliminated by the recursive
We thank Walter Doesschate of Amperex Electronic Corp., Julian
filter. However, in order to obtain multiplanar capability, to- Marshall of Xonics, Inc., and B. Douglas Lewis for comprehensive
mosynthesis algorithms would have to be used. It is not clear reviews and helpful criticism.
454 LEVIN ET AL. AJR:143, September 1984

REFERENCES Intracranial application of digital intravenous subtraction angiog-

raphy. Radiology 1982;144:529-534
1 . Price AR, James EA. Basic principles and instrumentation of
20. Gomes AS, Pals SO, Barbaric ZL. Digital subtraction angiogra-
digital radiography. In: Price AR, RO1IO FD, Monahan WG, James
phy in the evaluation of hypertension. AiR 1983;140:779-783
AE, eds. Digitairadiography: a focus on dilnicalutiity. New York:
21 . Ergun DL, Shaw CG, Kruger AA, Mistretta CA, Zamstorff WC.
Grune & Stratton, 1982:5-49
Video slow scan techreque for mask mode subtraction radiog-
2. Christensen EE, Curry TS Ill, Dowdy JE. Introduction to the
raphy. Presented at the annual meeting of the RadiOlOgical
physics of diagnostic radiology. Philadelphia: Lea & Febiger,
Society of North America, Dallas, November 1980
1978:234-248
22. Riederer SJ. Digital radiography: General Electric’s approach to
3. General Electric Medical Systems. Digital radiography. Milwau-
digital fluoroscopy and scanned projection imaging. In: Mistretta
kee: General Electric, 1980
CA, Crummy AB, Strother CM, Sackett JF, ads. Digital subtrac-
4. Cohen 0, Wagner LK, Aauschkolb EN. Evaluation of a digital
tion arteriography: an application of computerized fluoroscopy.
subtraction angiography unit. Radiology 1982;1 44:613-617
Chicago: Year Book Medical, 1982:142-150
5. Clark RA, Alexander ES. Digitial subtraction angiography of the
23. Ovitt 1W, Christenson PC, Fisher HD, et al. Intravenous angiog-
renal arteries: prospective comparison with conventional arterl-
raphy using digital video subtraction: x-ray imaging system. AiR
ography. Invest Radio! 1983;1 8:6-10
1980;135:1141 -1144
6. Meaney TF, Weinstein MA, Buonocore E, et al. Digital subtraction
24. Harnngton DP, Boxt LM, Murray PD. Digital subtraction angiog-
angiography of the human cardiovascular system. AiR
raphy: overview of technical principles. AiR 1982;139:781-786
1980;135:1153-1160
25. Brody WA. Digital radiography. New York: Raven, 1984:49-50
7. Crummy AB, Strother CM, Sackett JF, et al. Computerized
fluoroscopy:digital subtraction for intravenous angiography and
26. Brody WA. Hybrid subtraction for improved arteriography. Ra-
arteriography.AiR 1980;135:1131-1140 diology 1981;141 :828-831

8. Mistretta CA. TechnOlOgiCal considerations: equipment, imaging 27. Brody WA, Macovski A, PeIc NJ, Lehmann L, Joseph RA,
Edelheit LS. Intravenous arteriography using scanned projection
American Journal of Roentgenology 1984.143:447-454.

processing and subtraction techniques. In: Price AR, ROIIO FD,

Monahan WG, James AE Jr, eds. Digitalradiography: a focus on
radiography. Radiology 1981;141 :509-514

c!inica! utility. New York: Grune & Stratton, 1982:51-62

28. Van Lysel MS, Dobbins JT, Peppler WW, et al. Work in progress:
9. Flannigan BD, Gomes AS, Stambuk EC, Lois JF, Pals SO. Intra- hybrid temporal-energy subtraction in digital fluoroscopy. Ra-
arterial digital subtraction angiography: comparison with conven-
diology 1983;147:869-874
tional hepatic arteriography. Radio!ogy 1983;148:17-21
29. Foley WD, Keyes GS, Smith DF, et al. Work in progress: temporal
10. Ludwig JW, Verhoeven LM, Kersbergen JJ, Overtoom UC. energy hybrid subtraction in intravenous digital subtraction an-
Digital subtraction angiography of the pulmonary arteries for the giography. Radiology 1983;148:265-271
diagnosis of pulmonary embolism. Radiology 1983;147:639-645 30. Riederer SJ, Brody WA, Enzmann DA, Hall AL, Maler JK. Work
11 . CA, Crummy AB, Strother CM. Digital angiography: a
Mistretta in progress: application of temporal filtering techniques to hybrid
perspective. Radio!ogy 1981;139:273-276 subtraction in digital subtraction angiography. Radiology 1983;
12. Kruger RA. Basic physics of computerized fluoroscopy difference 147:859-862
imaging. In: Mistretta CA, Crummy AB, Strother CM, Sackett 31 . Aiederer SJ, Enzmann DA, Hall AR, PeIc NJ, Djang WT. Appli-
JF, eds. Digitalsubtraction arteriography: an application of corn- cation of matched filtering to x-ray exposure reduction in digital
puterized f!uoroscopy. Chicago: Year Book Medical, 1982:16- subtraction angiography: clinical results. Radiology 1983;
22 146:349-354
13. Enzmann DR, Djang WT, Riederer SJ, et al. Low-dose, high- 32. Aiederer SJ, Enzmann DR, Brody WA, Hall AL. AppliCatiOn of
frame-rate versus regular-dose, low-frame-rate digital subtrac- matched filtering to contrast material dose reduction in digital
tion angiography. Radiology 1983;146:669-676 subtraction angiography. Radiology 1983;147:853-858
14. Buonocore E, Pavllcek W, MOdiC MT, et al. Anatomic and func- 33. Kruger AA, Miller FJ, Nelson JA, Liu PY, Bateman W. Digital
tional imaging of congenital heart disease with digital subtraction subtraction angiography using a temporal bandpass filter: asso-
angiography. Radiology 1983;147:647-654 dated patient motion properties. Radiology 1983;145:315-320
15. Goodman PC, Brant-Zawadzki M. Digital subtraction pulmonary 34. Nelson JA, Miller FJ, Kruger RA, Liu PY, Bateman W. Digital
angiography. AIR 1982;139:305-309 subtraction angiography using a temporal bandpass filter: initial
16. Strother CM, Sackett JF, Crummy AB, et al. Clinical applications clinical results. Radiology 1982;145:309-31 3
of computerized fluoroscopy: the extracranial carotid arteries. 35. Aiederer SJ, Kruger AA. Intravenous digital subtraction: a sum-
Radiology 1980;136:781-783 mary of recent developments. Radiology 1983;147:633-638
17. Crummy AB, Strother CM, Lieberman AP, et al. Digital video 36. Gould AG. A digital fluoroscopic system using tandem video
subtraction angiography for evaluation of peripheral vascular processing units. In: Price AR, AoIIo FD, Monahan WG, James
disease. Radiology 1981;141 :33-37 AE, ads. Digital radiography: a focus on clinical utility. New York:
18. Guthaner DF, Wexler L, Enzmann DA, et al. Evaluation of pe- Grune & Stratton, 1982:81 -89
npheral vascular disease using digital subtraction angiography. 37. Kruger RA, Nelson JA, Ghosh-Roy D, Miller FJ, Anderson RE,
Radiology 1983;1 47:393-398 Liu P. Dynamic tomographic digital subtraction angiography us-
19. Carmody RF, Smith JRL, Seeger JF, Ovitt 1W, Capp MP. ing temporal filtration. Radiology 1983;147:863-867