Professional Documents
Culture Documents
1231–1238, 2001
Copyright © 2001 World Federation for Ultrasound in Medicine & Biology
Printed in the USA. All rights reserved
0301-5629/01/$–see front matter
PII: S0301-5629(01)00423-9
● Original Contribution
Abstract—Diagnostic ultrasonography has its well-established role in medicine. Nevertheless, the quanti-
tative characterisation of biological tissues by ultrasound (US) is still a main topic of research. Several
parameters have been explored with this purpose, (e.g. attenuation, backscatter coefficient, US speed). More
recently, mean scatterer space (MSS) has been proposed as a characterisation parameter. The objective of
this work was to investigate the potential of the singular spectrum analysis (SSA) to estimate MSS. This
method proposes the reconstruction of the periodic part of the original US signal from where the MSS of the
medium can be estimated. SSA is applied to simulated and real backscattered echoes from a phantom and
a bovine liver sample. Consistent results were obtained from both Monte-Carlo simulation and real data.
They were compared with literature. Presently, precision, accuracy and sensibility of SSA are being
investigated. (E-mail: wagner@peb.ufrj.br) © 2001 World Federation for Ultrasound in Medicine &
Biology.
Key Words: Singular spectrum analysis, Ultrasound, Tissue characterisation, Mean scatterer space.
1231
1232 Ultrasound in Medicine and Biology Volume 27, Number 9, 2001
duced a characterisation of the scatterer distribution of each position of the window, the M data points inside it
tissue microstructures with a technique called spectral can be described by a linear combination of the vectors
autocorrelation (SAC). One of their results is the estima- Ek, according to the following equation:
tion of MSS from 2-D maps. The main difference be-
冘a E , 1 ⱕ j ⱕ M,
tween the classical power spectrum and the spectral M
autocorrelation is that the SAC function includes the x i⫹j ⫽ k
j
k
(1)
phase differences between different spectral components. k⫽1
Simon et al. (1997) proposed a computationally
efficient algorithm for MSS estimation. They used an where the coefficients akj are called the principal com-
analysis based on the spectral redundancy present in ponents (PCs).
backscattered echo signals and generated spectral lines Each data window can be understood as one single
through a quadratic transformation of the RF echo signal. point in a M-dimensional phase space. The singular
The main advantage of this approach is the possibility of vectors (Ek) then represent the principal axes (bases) of
real-time implementation and superior results when com- this phase space. Each singular vector (eigenvector)
pared with amplitude-only algorithms. When compared points toward the directions with most energy (following
with amplitude-phase techniques, their results are noted an optimisation criterion); thus, the projection of the
as comparable. original points onto them is the most efficient for that
The basic assumption that supports the concept of specific data set.
MSS estimation is that the periodic and nonperiodic Two important results from the eigenstructure of
structures can be separated in their corresponding US principal component analysis can be summarised:
backscattered echoes. This separation can be repre- 1. Eigenvectors of the correlation matrix Cx pertaining
sented, for example, by the discrete (“fence-like” well- to a zero-mean random vector x define unit orthog-
defined frequencies) and continuous part of the signal onal vectors representing the principal directions
spectrum, respectively. along which the statistical variance has its extreme
From this point of view, singular spectrum analysis values; and
(SSA) seems to be a powerful tool for time-series anal- 2. Each associated eigenvalue defines the extreme
ysis, different from the common approaches and not yet value of the variance in that direction. The total
well explored for biological tissue characterisation. It has variance of the time series is divided among the M
the capability to decompose periodic and aperiodic struc- different basis.
tures from the time signal, even in a noisy ambient. PCA and SSA have the same mathematical descrip-
The main objective of this work was to investigate tion. PCA has been applied to data compression, chang-
the potential of SSA to estimate MSS when applied to ing the original data set to a new one based on the highest
simulated and real backscattered echoes from phantoms variance information described by the principal axes (the
and bovine liver. Results from a Monte-Carlo simulation, variance is associated with the eigenvalues.) SSA uses
from an experimental phantom, as well as from a bovine information from eigenvalues/eigenvectors to describe
liver sample, are presented and discussed. particularities present in time series. In our approach, we
looked for the basic periodic behaviour from the data,
and it may not necessarily be described by the highest
THEORY
variance components.
SSA is a tool commonly used in nonlinear dynamics Vautard and Ghil (1989) showed that the presence
studies (Broomhead and King 1986; Vautard and Ghil of eigenvalues with values close enough to be identified
1989; Fowler and Kember 1998) and in several signal- as a “pair” could be associated with a periodic compo-
processing areas (Therrien 1992). It is based on the nent of the signal. The frequency of this periodic com-
classical principal component analysis (PCA), where a ponent can be found by taking the average of the peak
time-series data vector x with length N-points can be value of the amplitude spectrum of the two associated
expanded with respect to a complete orthonormal basis eigenvectors. After we can identify the eigenvalue pairs
composed of M vectors (Ek, 1 ⱕ k ⱕ M). present in the data, it is possible to reconstruct and
The vectors Ek are called eigenvectors and are ob- characterise separately the periodic and nonperiodic
tained from the diagonalisation of the correlation matrix structures of the data just by selecting the correct eigen-
(Cx) from the original N-data points. This matrix is vectors. In the same way, this decomposition can opti-
constructed by moving a window of M points (M ⬍ N) mally capture certain parts of signal behaviour, selecting
along the sequence x. This window is shifted by one a correct subset of bases from all bases (Vautard et al.
point at a time. The position of the window is indicated 1992).
by the index “i”, so, “i” varies from 1 to N ⫺ M ⫹ 1. For Some drawbacks with this technique have to be
Singular spectrum analysis on US simulated signals ● W. C. A. PEREIRA and C. D. MACIEL 1233
The figure of merit adopted for this Monte-Carlo assumed c ⫽ 1500 m/s at 26°C. The second peak is in the
analysis is depicted in Figs. 8 and 9 and is the same cited first harmonic of the previous peak (0.508 MHz).
in Simon et al. (1997). It was chosen as such to ease the Figure 7 shows different levels of jitter (a ⫽ 5%;
comparison of the results. There, one can find the per- b ⫽ 10%; c ⫽ 20%; d ⫽ 30%) to the same SNR (near 0
formance curves for five different levels of SNR, as a dB). Two results are important in these graphics. The
function of jitter. In Fig. 8, a correct estimate is defined first one is that some estimates are grouped around the
as the one that falls within the range of 5% of the correct first harmonic of the simulated original frequency. The
value. In Fig. 9, this range is set to 25% of the correct second result is that, as jitter becomes higher, the esti-
value. The interpretation of them is in the next section. mates start to spread around the original clusters. Each
The US bovine liver samples were processed with item of the figure has a clustering plot (top) and the
the SSA algorithm proposed, and the final result is pre- respective histogram (bottom). The clustering is com-
sented on a histogram in Fig. 10. Each histogram class posed of the MSS estimates (y-axis is the SNR level and
x-axis is the MSS frequency). For the histograms, the
has a width of 20 kHz. This figure shows a maximum of
x-axis is the same (MSS axis) and the y-axis the number
the histogram in 0.20 MHz (4.0 mm) and a range be-
of events (estimates). For this analysis, the size of the
tween 0.19 MHz (4.8 mm) and 0.22 MHz (3.1 mm). The
window (M) is 256 samples.
MSS value, as well as its range, were estimated from the
In our work, we made Monte-Carlo analysis with
relation MSS ⫽ c/(2*MSS_Frequency), where c is the
different levels of SNR and jitter, as summarised in Figs.
US speed in the medium which, for this experiment, was 8 and 9. It is possible to observe the consistency of the
estimated as 1630 m/s at 20°C (Pereira et al. 2000). The method: lower jitter and low SNR generate a high per-
parameter MSS_frequency is obtained as the correspond- centage of correct estimates. In Fig. 9, there is an inter-
ing frequency of the amplitude peak of the FFT (fast esting aspect; for almost all the curves, the best perfor-
Fourier transform) performed on the SSA reconstructed mance is not at jitter equal to zero, but only when it is 5%
US signal. or 10%. One possible explanation for this would be that,
for low jitter, some estimates are locked in the harmonics
DISCUSSION and, as jitter increases, the cluster spreads and some
estimates may have returned to be around the correct
For the results with the phantom signals, it is im- value. This behaviour can also be observed in Fig. 8
portant to notice that the peak in Fig. 6b is in 0.254 MHz, (lines for ⫺3 dB and 5 dB). In this study, we have made
which corresponds to a periodicity of 2.95 mm (the real an analysis to quantify the repeatability of the process. If
spacing is 3 mm). This error seems to be acceptable, we assume 50% of correct estimates as a sign of consis-
considering the US speed in water was not estimated but tency, we can see that, for the curve with SNR of 12 dB,
1238 Ultrasound in Medicine and Biology Volume 27, Number 9, 2001
jitter goes as far as 17%, approximately. Even for a SNR For this work, the next step is to devise a method-
as low as 0 dB, which is a severe adverse condition, the ology to optimise the separation between periodic and
process tends to generate consistent estimates up to aperiodic subspaces, so that one can extract an unique
around 14% of jitter (Fig. 8). The performance of the periodic signature of the medium. The precision, accu-
process continues to decrease as jitter goes up. It is racy and sensibility of SSA are also being investigated
important to have in mind the question of until what level with a more significant amount of experimental data.
of jitter a periodic structure can be understood as such. After that is reached, we can try to use SSA automati-
Simon et al. (1997) presented a comparison of dif- cally to localise intermittent oscillations.
ferent types of MSS algorithms. The results shown in
Acknowledgement—To the Brazilian agencies CNPq and CAPES.
Figs. 8 and 9 are similar to those in their article (Fig. 11c
and d). To generate the Monte-Carlo simulation, they
adopted a parameter Ad, defined as the amplitude rate REFERENCES
between the diffuse and regular scatterers, which is Allen MR, Robertson AW. Distinguishing modulated oscillations from
coloured noise multivariate datasets. Climate Dynam 1996;12:775–
equivalent to a SNR value of 20 dB in our definition. In 784.
a first approach, consider the line of 12 dB of Figs. 8 and Allen MR, Smith L. Monte Carlo SSA: Detecting irregular oscillations
9 to compare with the results of Simon et al. (1997). in the presence of coloured noise. J Climate 1996;9:3373–3404.
Bridal SL, Saı̈ed A, Chérin E, Lefèbvre F, Laugier P, Berger G.
Their results at 15% of jitter indicate a performance High-resolution quantitative echography with tissue parameters. J
inferior to 50%, but our results are near 60%. From 5% Echograph Méd Ultrasons/JEMU. 1998;19(2/3):204 –211.
to 15% of jitter, our results are near 10% higher than Broomhead DS, King G. Extracting qualitative dynamics from exper-
imental data. Physica D 1986;20:217–236.
theirs. In the range of 25%, the results are similar. Fellingham L, Sommer F. Ultrasonic characterization of tissue struc-
The simulation model used assumes that all the ture in the in vivo human liver and spleen. IEEE Trans Sonics
scatterers are in one dimension (1-D). When using real Ultrason 1984;SU-31:418 – 428.
Fowler AC, Kember G. Singular systems analysis as a moving-window
US signals, all the scatterers that are within the US beam spectral method. Eur J Appl Math 1998;8:55–79.
contribute to the signal. In other words, the scatterers are Kadah YM, Farag AA, Zurada JM, Badaw AM, Yossef AM. Classifi-
in a 3-D distribution. So, for the signals from bovine cation algorithms for quantitative tissue characterization of diffuse
liver disease from ultrasound images. IEEE Trans Med Imaging
liver, we believe that the fact that SSA could deal with 1996;15(4):466 – 478.
this 3-D distribution and make the estimates converge to Landini L, Verrazzani L. Spectral characterization of tissue microstruc-
a limited range with a clear peak is, in itself, an indica- ture by ultrasound: A stochastic approach. IEEE Trans UFFC
1990;37:448 – 456.
tion of the potential of the SSA analysis. The range in Narayanan VM, Molthen RC, Shankar PM, Vergara L, Reid JM.
which the estimates were 3.1 to 4.8 mm is also compat- Studies on ultrasonic scattering from quasi-periodic structures.
ible with the liver microstructure, and depends on how IEEE Trans UFFC 1997;44:114 –124.
Palus M, Dvorak I. Singular value decomposition in attractor recon-
the sample was cut and exposed to the US beam (i.e., the struction: Pitfalls and precautions. Physica D 1992;58:221–234.
orientation of the lobes). Pereira FR, Pereira WCA, Machado JC. Ultrasonic wave speed mea-
surement using the time delay profile of RF backscattered signals.
CONCLUSION Ultrasonics 2000;38:708 –710.
Simon C, Shen J, Seip R, Ebbini, ES. A robust and computationally
A study is presented on the application of the SSA efficient algorithm for mean scatterer spacing estimation. IEEE
Trans UFFC 1997;44(4):882– 894.
to US backscattered signals from a simple phantom and Therrien C. Discrete random signals and statistical signal processing.
a bovine liver sample to estimate the mean scatterer Englewood Cliffs, NJ. Prentice Hall Processing Series, 1992.
spacing. In both cases, it was possible to obtain consis- Varghese T, Donohue KD. Characterization of tissue microstructure
scatterer distribution with spectral correlation. Ultrason Imaging
tent estimates of the MSS. A Monte-Carlo analysis of the 1993;15:238 –254.
estimates of MSS based on the SSA was performed on Varghese T, Donohue KD. Mean scatterer spacing estimate with spec-
simulated signals from a simple 1-D model from the tral correlation. J Acoust Soc Am 1994;96:3504 –3515.
Varghese T, Donohue KD. Estimating mean scatterer spacing with the
literature, which has shown the behaviour of SSA in the frequency smoothed spectral autocorrelation function. IEEE Trans
presence of different levels of jitter and SNR. UFFC 1995;42:451– 463.
SSA was shown to be a powerful way to estimate Vautard R, Ghil M. Singular-spectrum analysis in nonlinear dynamics
with applications to paleoclimatic time series. Physica D 1989;35:
MSS, even compared to the algorithm published in the 395– 424.
literature. In addition, SSA opens the possibility of Vautard R, Yoiu P, Ghil M. Singular-spectrum analysis: A toolkit for
studying the decomposition of US signals (periodic and short, noisy chaotic signals. Physica D 1992;58:95–126.
Wear K, Wagner R, Insana M, Hall T. Application of autoregressive
aperiodic subspaces) for a possible characterisation of spectral analysis to cepstral estimation of mean scatterer. IEEE
biomedical US data. Trans UFFC 1993;40:50 –58.