Тhe envelope correlation spectrum algorithm as a method for ultrasonic tissue characterization тема автореферата и диссертации по физике, 01.04.06 ВАК РФ

V t w

MOSCOW STATE UNIVERSITY PHYSICS FACULTY

Manuscript

JACQUELINE MARIE CLEMENCE GALLET

THE ENVELOPE CORRELATION SPECTRUM ALGORITHM AS A METHOD FOR ULTRASONIC TISSUE CHARACTERIZATION

01.04.06 - Acoustic

Summary of the Dissertation submitted for the degree of DOCTOR OF PHILOSOPHY

Scientific Advisor J P Jones .Professor, PhD

Moscow 1993

Работа выполнена в Калифорнийском Университете на факультете радиологии

Научный руководитель: доктор философии, профессор Дж.П.Джонс

Официальные оппоненты: доктор физ.-нат. наук В.А.Буров доктор техн. наук Л.Р.Гаврилов

Ведущая организация: Институт проблем управления РАН

Защита состоится " июня 1993 г. в 15

час. 30 мин. на заседании Специализированного Совета К.053.05.92. отделения радиофизики физического факультета ЯГУ им. М.В.Ломоносова в ауд. 5-19. По адресу: 119899, Москва, Ленинские горы, МГУ, физический факультет.

С диссертацией можно познакомиться в библиотеке физического факультета МГУ.

Автореферат разослан: мая 1993 г.

Ученый секретарь

Специализированного

Совета

И.В.Лебедева

SYNOPSIS

Ultrasound has been used as a tool in diagnostic medicine for over three decades. Although the technology has evolved slowly, it has always striven to achieve diagnostic effectiveness.

The acoustical frequencies chosen for diagnostic

medicine range from 1 MHz to 20 MHz and represent a trade-off

between spatial axial and lateral resolutions and the frequency

dependent attenuation of tissue. Also, intensity levels are usually 2

below 100 mW/cm [spatial peak temporal average intensity]. Within these parameters no adverse effects have as yet been demonstrated in human.

The frequency range of 1 to 20 MHz corresponds to wavelengths of the order of 1 mm to 0.1 mm. Even though these wavelengths correspond to rather small physical dimensions, the spatial axial resolution for an acoustical system is roughly defined as being four times the absolute wavelength. Therefore, in an ultrasonic imaging system, two anatomical structures will be axially resolved if they are separated in distance by 4 to 0.4 mm at these frequencies. The lateral resolution, which corresponds to the ability of the imaging system to differentiate two structures which lie on an axis

perpendicular to the axis of emission, is not as well defined as the axial resolution as it strongly depends on the physical dimensions of the transducer. The general rule of thumb is that the lateral resolution of an acoustical imaging system can be approximated as being four times the axial resolution. Even though, within these defining parameters, gross anatomical features can be observed, the main problem of present ultrasonic imaging lies in the fact that the gray scale image produced is unsharp. This is analogous to the image a myopic sees when looking, without corrective lenses, at an object beyond the focal zone of their eyes. This image unsharpness has been defined as speckle and is due to the basic physical nature of acoustical waves interacting with a biological medium whose structure is on the order of or smaller than the interrogating wave length.

When a wave is reflected from a specular reflector, it can be described in one dimension as R[z] with an amplitude, IAI, and a phase, z. Thus,

R(z) = A eikz

(1.1

If there exists only one reflection plane (Figure 1.1] the phase will be the same for all reflected waves as the parameter z is a constant. When the magnitude of Equation 1.1 is then calculated [Equation 1.2), the phase part is canceled. Thus,

,2 I -t I2 R(z) = A e I

(1.2)

= (A eikz) (A e"ikz)

= A

If, on the other hand, several reflection planes exist, each plane will contribute a different phase to the reflected wave equation [Figure 1.2] as the z distance is no longer a constant. Therefore, since waves add constructively or destructively, the resultant sum at each point of interrogation will contain a phase part which will not necessarily cancel when the magnitude is taken. This, in essence, can be described as a random process. Some algorithms have been proposed to reduce or even eliminate speckle but these have been relatively unsuccessful. This randomness does not imply that the physical and structural information of the interrogated object is lost. On the contrary, it is recognized that the tissue properties are imbedded in the detected reflected acoustical signal. The main thrust of recent

Figure 1.1 Representative ray diagram depicting a transducer emitting waves being reflected at right angles from a specular reflector.

M t

z,2 Z3

11

Figure 1.2 Representative ray diagram depicting the effect of phase. Each layer contributes a different phase value to the final reflected

studies has been to extract this tissue information from the signal by digital signal processing techniques.

There is a large amount of information, generally unused by conventional ultrasound imaging systems which can be extracted from the reflected acoustical signal. This information extraction process can be categorized into three general approaches: dynamic characterization, parameter estimation and structure characterization. Parameter estimation, as the name implies, tries to determine through some form of measurement or estimation technique acoustical properties such as attenuation, impedance and velocity. A vast compilation of bio-acoustical data exists which can be compared to in-vivo results. The experimental methodology is fairly well developed at transducer frequencies of less than 100 MHz. The interpretation of the results which is based on the interaction of the physical properties of the acoustical wave with the biological properties of the insonified target is perhaps the major problem associated with these results . Dynamic characterization investigates the evolution of an insonified target as a function of time. Since these types of measurements are usually lengthy, such as the ones examining foetal development, interest has not to date been sustained. The last category, structure characterization, attempts to determine a spatial dimensional signature from the detected backscattered ultrasonic waves which are also known as A-lines. Various signal processing techniques are applied to the detected A-line in order to derive a unique quantitative signature. Ideally, the algorithmic method is easy to implement in a

ilinical setting.

In this present research, we will investigate a means of ibtaining useful and unique information from an interrogated target by he method of structure characterization. Based on some preliminary xperiments [Jones, JP and Kovack.R (1980) A computerized data nalysis system for ultrasonic tissue characterization. Acoustical maging 9, 503-512] we will largely utilize an algorithm known as the nvelope correlation spectrum referred to by its acronym throughout lis manuscript as ECS. The ECS is different from previously uggested structure characterization algorithms in that it does not epend on the distribution of the scattering centers within an isonified object, nor do the effects of attenuation need to be ansidered in the analysis.

It is well known that tissue exhibits a structural identity, has also been observed that infiltrating disease tends to alter that iherent structure. Ideally, a structure characterization algorithm ould in its process reveal that the tissue under scrutiny is no longer Drmal but has an abnormality. Depending upon the refinements of e algorithm, it should also be able to quantify the degree to which e disease process has advanced.

The ECS algorithm is based upon the concept that an terrogating ultrasonic pulse of Gaussian shape retains its overall ape while interacting with frequency dependent attenuating tissue,

lthough suffering a shift in center frequency and a pulse broadening . : the initial pulse can be described as

y(t) = exp(-a2t2)cos(27tfQt) (1.3)

where a is the standard deviation and f is the mean frequency

hen it can be shown that the shape of the propagated pulse through a issipative medium is

Y(f) = B exp -

where B is a constant 0 is the new standard deviation

f is the new mean frequency

Much of the current research involves computer lations. For these simulations the attenuation of the medium was onsidered since the effects of attenuation should not affect an jpriate structure algorithm and since the effects of attenuation >e added later, if necessary. This reduced the complexity of the uter programming and enabled us to compare three different ture algorithms on an equal basis. We also hypothesize that the ope correlation spectrum algorithm is independent of uation. That is, the structural information of the scatterers can mputed by the algorithm irrespective of the inherent frequency idence of the tissue under investigation. In this case, the above sian representation as described by Equation 1.4 would remain at lentical center frequency and standard deviation.

In addition, the simulated data were also evaluated by two structure algorithms for tissue characterization and compared to CS. The first was the histogram of the first peak of the orrelation of the power spectrum or HIP [Joynt.L (1979) A astic approach to ultrasonic tissue characterization. Stanford ronics Laboratory, Technical Report No. G557-4]; the second was dimensional textual algorithm first presented by Gonzalez and searchers [Gonzalez, V; Jones, JP; Ferrari, L; and Behrens, M ) The analysis of A-mode ultrasound waveforms with a one isional texture algorithm. IEEE Ultrasonics Proceedings.]. It was red that either of these two algorithms would contribute greatly ther the knowledge base of tissue characterization during the

preceding decade. Unfortunately, neither have withstood the test of time as both staffer to a certain degree from lack of a basic understanding of the physical principles involved in the algorithmic procedures. Chapter 2 entitled Tissue Characterization in General of the doctoral thesis more fully explains the differences of the three ultrasonic tissue characterization algorithm chosen for the comparison.

As a consequence, the three algorithms, that is the ECjS, the one dimensional texture algorithm and the HIP were extensively tested to determine their dependence on data window length, Fourie transform processing length and simulated input data. The purpose < the computer simulations was to determine the limitations of each algorithm and to determine as well the algorithm which did not deviate strongly from the expected values.

It was also a concern that the characterization of tissue would perhaps have to be restricted to Gaussian profile pulses. If ultrasonic tissue characterization is to be successful in the future, the the structure algorithms developed must keep pace with available technology and clinical useage. Therefore, in order to further test th three structure algorithms, a non-Gaussian pulse profile was utilized ; the incident ultrasound pulse. This type of pulse profile was chosen : many current medical transducers generate pulses which are non-Gaussian in shape. The results of the computer simulations are detailed more completely in Chapter 3 entitled Computer Simulation

lie thesis.

Even though an algorithm can stand up to extensive nputer simulations, it does not necessarily imply that it will survive test of a few simple structured test objects. Computer simulations 1 only be expected to yield a general overview of the rigidity of an orithm. It can not anticipate each and every factor and variable of actual experiment. Therefore, two test objects of differing physical iperties were chosen to yet further investigate the rigidity of the S algorithm in relation to the other two, namely the HIP and the ; dimensional texture algorithm.

The two test objects selected were first an articulated m and secondly a segment of bovine liver. Each were set up in a ter tank in a classical acoustical experiment. The sponge was )sen as it is composed of holes and fibers randomly arranged. The ults of the algorithms for the numerical size of the scattering iters can also be checked physically by taking a graduated rule and asuring the size of the scattering centers. The bovine liver was ked as a preliminary test of the ECS algorithm on biological nples. It is an organ which is often chosen in ultrasound leriments as its physical structure can be described as a regular ay of scattering centers in its normal state. This particular piece of Ine liver also had a venal hole of a physically measurable dimension ich the ECS algorithm essentially selects from the data along with : most probable lobule size. It was not expected that the HIP would

be capable of duplicating the results of the ECS because of the fundamental basis of the HIP algorithm . The HIP algorithm only detects the average size of the scattering centers of the interrogated test object, and not individual sizes of the many components of the test object. On the other hand, the one dimensional texture algorithm should yield different pictorial representations for the sponge as compared to the bovine liver. Chapter 4 and 5 of the thesis investigates the experimental differences between the three specified algorithms, that is, the one dimensional texture algorithm, the histogram of the first peak of the autocorrelation of the power spectrum and the envelope correlation spectrum between the two tes1 objects, the bovine liver and the articulated foam.

It will be shown in this present research that the ECS algorithm is a far superior method for ultrasonic echo wave-form analysis than the HIP or the one dimensional texture algorithms [cf Figures 1.3, 1.4 and Table 1.1]. The extensive computer simulations with Gaussian and non-Gaussian pulse profiles were extremely robust. The experimental data from the two test objects stood up either to previously published data in the case of the bovine liver or to subsequent physical measurements in the case of the articulated foam.

100 - Envelope Correlation Spectrum

80 - , Rayleigh Fit

60 - \ \ ^ 0.31cm

40 - \

20 -0 H >— —i i— —i-- ■ *

O 100 200 300 400

Point Number (x 2440 Hz)

î 1.3 Rayleigh distribution fit to envelope correlation spectrum result iraging collected data set from the articulated foam. This results in a spacing of approximately 1.3 mm which was comparable to physically iring the spacings of the articulated foam with a rule.

FFT Length Scattering Dimension

(Points) (mm)

64 1.2

128 1.6

256 3.2

512 6.3

1024 6.3

8192 8.0

Table 1.1 Variation of sponge scatterer size as a function of Fouri transform length. It is quite evident from the apparent fluctuation depend on the Fourier transform length, that the information obta the histogram of the first peak of the autocorrelation of the power is of limited use.

*ure 1.4 One dimensional texture algorithm result from the analysis the bovine liver as compared to the articulated foam. Even though e slope of each is different no further quantitative information can be tained. This is believed to be less than one can obtain from a tditional ultrasonogram.

REFERENCES

1. Jones JP, Chandraratna PAN, Tak T, Kaiser S, Yigiter E and C J (1989) Detection of early fatty plaque using quantitative ultrasou methods. Abstract presented at the 18th International Symposiu Acoustical Imaging, Santa Barbara, California September 18-20.

2. Gallet JMC and Jones JP (1991) The envelope correlation spectrum algorithm as a method for ultrasonic tissue characteriza Abstract presented at the 16th International Symposium on Ultrasonic Imaging and Tissue Characterization, Arlington, Virgin: June 3-5.

3. Gallet JMC and Jones JP (1991) Quantitative characterizatic tissue structure using the envelope correlation spectrum algorithr Abstract presented at the World Federation of Ultrasound in Biolc and Medicine, Copenhagen, Denmark, September 2-6.

4. Jones JP, Chandraratna PAN, Tak T, Kaiser S, Yigiter E and Gallet J (1991) Detection of early fatty plaque using quantitative ultrasound methods. In Acoustical Imaging, Volume 18, Edited b; Lee and G Wade, Plenum Press, New York, 1-6.

5. Jones JP and Gallet JMC (1992) Attenuation measurements 100 to 800 MHz with the acoustical microscope. Abstract presei at the 36th annual convention of the American Institute of Ultras^

Medicine, San Diego. California.

Gallet JMC and Jones JP (1992) Tissue characterization by the thod of the envelope correlation spectrum. Abstract presented at

1992 IEEE Ultrasonics Symposium, Tuscon, Arizona, October 20-

Gallet JMC and Jones JP (1993) Tissue characterization by the thod of the envelope correlation spectrum. Accepted for Dlication in the 1992 IEEE Ultrasonics Symposium Proceeding's.

Gallet JMC and Jones JP (1993) Ultrasonic tissue characterization the method of the envelope correlation spectrum algorithm. To be alished in Ultrasound in Medicine and Biology.

Chandraratna PAN, Choudhary S, Jones JP Chandrasoma P Kapoor _nd Gallet JMC (1993) Visualizations of myocardial cellular hitecture using acoustical microscopy. In American Heart Journal,

выводы

1. Показано, что алгоритм спектральной корреляции огибающей быстро сходится, как функция размера окна Фурье-преобразовате-ния. Отсюда следует, что даже для малого окна этот алгоритм может выделить средние размеры рассеивателей.

2. Показано, что этот алгоритм является относительно несложным алгоритмом при проведении клинических исследований. При этом данные, получаемые при помощи работающих в реальном режиме времени ультразвуковых сканеров, могут быть либо немедленно обработаны при помощи этого алгоритма, либо сохранены для более позднего последующего анализа.

3. Найдено, что этот алгоритм является независимым от начального статистического распределения рассеивателей. Например, регулярность структуры рассеивателей соответствует основному пику амплитуды в частотной области Фурье-преобразования. Также, для гауссовского распределения рассеивателей, среднее значение размеров элементарных рассеивателей сохраняется после применения этого алгоритма.

4. Показано, что этот алгоритм является более надежным и устойчивым в использовании по сравнению с прежде используемыми алгоритмами характеризации тканей, основанными на положении первого пика автокорреляционной функции спектра мощности или одномерном текстурном анализе.