Global breast attenuation: Control group and benign breast diseases

ULTRASONIC

IMAGING

12,

47-57,

(1990)

GLOBAL BREAST ATTENUATION : CONTROL AND BENIGN BREAST DISEASES

G. Berger I, P. Laugier l, J.C. Thalabard

GROUP

2 and J. Perrin 1

1 Lab. Biophysics -UA CNRS 593 - Cochin University-Hospital 24, rue Fbg St.-Jacques, 75674 Paris, France * INSERM - U292- Bidtre Hospital 78, ave. du General Leclerc, 94275 Le Kremlin-BicBtre,

France

This paper deals with the estimation of the slope of attenuation in human breast tissue. The measurement is done in the reflection mode with a short time Fourier analysis. All important factors such as diffraction effect, tissue depth and specular reflectors are taken into account. A population of 49 normal women shows large inter-individual variations of the attenuation coefficient. A multiple linear regression allows correlation of this variation with the duration of the woman’s genital life and pregnancies. A preliminary study is done on 10 benign diffuse breast diseases and shows a weak correlation with the type of The utility of the quantification is the breast : normal or pathological. discussed for one case of large fibrocystic disease. 01990 Academic Press,Inc. Key

words

: Attenuation ; benign breast disease ; human multiple linear regression ; reflection mode .

breast

;

INTRODUCTION From a general standpoint, breast disease is a very important clinical area in which classical echography is widely used in routine practice. Different approaches have been taken in the field of tissue characterization, using texture analysis [1,2], or an estimation of acoustic tissue parameters [3-51. Attenuation has been estimated in the transmission mode and a tomographic technique allows for breast attenuation images as well as for quantification [3,5]. Nevertheless, in tomographic the representation, the frequency dependence of attenuation is not given. The ability to measure this dependence with the slope of attenuation (dB/cm.MHz) in the reflection mode has been

47

0161-7346/90 $3.00 Copyright 0 1990 by Academic Press, Inc. All rights of reproduction in any form reserved.

BERGER ET AL,

demonstrated by Parker [4] on a normal human breast. On the other hand, such a measurement of the attenuation in pulse echo mode does not require a totally specific designed system as the one needed in the tomographic mode. It can be performed by using a conventional B-scan connected to a microcomputer. Our group is involved in attenuation estimation (dB/cm.MHz) [6] and we extended this technique to breast investigation. Our interest in this clinical study was benign breast disease. There appears to be agreement concerning the cancer risk associated with this pathology [7]. On the other hand, iterative biopsies of the gland appear to increase the associated cancer risk [8]. The authors introduce attenuation estimated in the reflection mode as a global breast index which could characterize breast tissue and be applied to the investigation of benign breast diseases. This index could allow the followup of fibrocystic disease as well as their evolution under therapy, which is necessary to reduce the cancer risk. The stability of the estimation with respect to intra-individual variability has been assessed by repetitive estimations on the same volunteer. It is clearly necessary to begin with a study on a control group in order to estimate the influence of inter-individual variations. Before we present the conclusions we drew both from this group and from some pathological cases, we will describe our methodology.

MEIH0DoLoGY The technique we used for attenuation estimation is an extention of the technique we have developped for in vivo muscle investigation [9-l 11. Two algorithms based on a short time Fourier analysis were : evolution of the centroid versus depth [ 121 systematically employed and evolution of the attenuation coefficient (dB/cm) versus frequency [131. The experimental setup consisted of a mechanical sector scanner (5 MHz focused monoprobe, 19 mm in diameter, 4 MHz bandwidth) which allowed localisation of the mammary gland. We selected signals coming only from the breast tissue with an adequate time delay. For each breast we acquired 256 A-lines with a sampling rate of 25 MHz. Those 256 A-lines are divided into 8 independent sectors, each of 32 scanned with a one lines. Within a sector, lines are approximately degree step. Thus, adjacent A-lines in a sector are probably partly correlated, which has no consequence on the precision of the estimation. The same variance could probably be reached with a smaller number of

48

GLOBAL

BREAST

ATTENUATION

A-lines. The tissue depth available is an important parameter which will be discussed. The other considerations to be taken into account when dealing with attenuation estimation are the specular reflector noise and the diffraction effect. These aspects have already been discussed elsewhere [11,14,15]. In such a technique, the tissue thickness “D” will dramatically influence the variance of the estimator, which is inversely proportional to the third power of D [ 10,161. Increasing the thickness yields a greater improvement in precision than does increasing the number of A-lines. The dorsal decubitus position, which is the one used in routine thickness along the echography of the breast, reduces the gland propagation axis. The average available tissue depth appears to be less than 2 cm. For this reason, we chose to perform the acquisition in a sitting position. The corresponding presentation of the glandular tissue gives a minimum of 2.5 cm depth along the propagation axis. In some cases, depending on the morphology, both presentations give a gland thickness of 3 cm. In this case, the different attenuation curves are well superimposed (Fig. 1). The relative precision of each estimation under these considerations is about 510%. We have already pointed out the problem for attenuation estimation due to specular reflectors [9]. This problem appears to be less critical in breast than it is in muscle. Breast is a relatively more homogeneous tissue. In most cases, the evolution of the averaged centroid versus depth appears quite regular (Fig. 2).

Fig. 1 2.0

3.0

4.0

6.0 5.0 FREQUENCYWHZ)

49

Normal human breast: attenuation curves versus frequency for left breast decubitus (continuous line), left breast sitting (dotted line) and right breast sitting (dashed line).

BERGER ET AL

26 e g

Fig. 2

d

5. I2

10.24

15.36

20.48 25.60 DBPTB p’

30.72

All these considerations influence of the frequency-dependent attenuation account before starting the investigation

BREAST ATTENUATION

IN A CONTROL

Example of the centroid and spectral variance evolutions versus tissue depth in a normal human breast tissue : right breast (continuous line) and left breast (dotted line).

the validity of the estimation . They have to be taken into of the control group.

GROUP

The control group was composed of 49 normal healthy women with ages 19 to 65. Both breasts were systematically explored. Both values were generally in good agreement, providing a global attenuation index. The attenuation distribution in the control group is presented in figure 3. We note very large inter-individual variations. Values range from 0.8 to 2.91 dB/cm MHz : the mean and standard-deviation are ( Eq. (1)) : p = 1.68 +

0.48

dB/cm

MHz

(1)

This standard deviation is large compared to the inter-individual which deal with the attenuation variations already published distribution in different populations and different tissues (liver, muscle). For comparison, we recall the standard deviation we found on muscle within a population of 30 healthy volunteers : 0.07 dB/cm MHz as well as the standard deviation we found on muscle within a population of 45 carriers of Duchenne Muscular Dystrophy : 0.09 dB/cm MHz [ 10,181. The comparison can be made because those values result from the same laboratory using the same technique.

50

GLOBAL BREAST ATTENUATION

0

9-

c C u R R E N c sE

“i 7t

6t 5j

4t 3

2 I -.11)1 0

0.65

1.25

1.85

2.45

3.05

Attenuation (dB/cm MHz) Fig. 3

Attenuation

distribution

in a population

of 49 normal human volunteers.

We tried to understand the origin of this variation in the group of 49 healthy volunteers with a uniform age distribution. As a starting point, we observed different evolutions of the attenuation coefficient versus frequency according to the age of the woman. Figures 4 and 5 present respectively the breast attenuation curves of a woman aged 25 and of a woman aged 53. Note the similarity of the curves corresponding to the right and left breasts. In preference to the woman’s age, we choose the duration of the genital life, defined as the woman’s age minus the age of her first menstruation. Genital life duration rather than age reflects breast tissue maturation. The previous observation indicates that the breast attenuation tends to decrease with

25 years old Fig. 4

51

Attenuation curves of the right (continuous line) and left (dotted line) breasts of a healthy woman aged 25.

BERGER

ET

AL

53 years old Fig. 5 2.0

3.0

4.0

5.0 60 FREQUENCY blH2)

the duration of the genital in this paper.

life.

This

Attenuation (continuous line) breasts aged 53.

conclusion

will

curves of right line) and left (dotted of a healthy woman

be quantified

A more complete correlation with physiological parameters has been performed using a multiple linear considered the following parameters :

later

and acquired regression. We

1. Duration of the genital life : defined as the woman’s age minus the age of her first menstruation. : the woman had no (= 0) or one or more (=l) 2. Pregnancies pregnancies. 3. Phase : the phase is divided into four categories (1 = follicular phase, 2 = luteal phase, 3 = oral contraceptive, 4 = menopausal phase). 4. Family breast history : a history of breast cancer is not found (= 0) grandmother, aunt) or found (= 1) for the relatives (mother, 5. Personal breast history (whether benign or not): exists (= 0) or does not exist (= 1) Variables under study are : 1. Attenuation 2. Logarithm

of the attenuation

The correlation matrix (table I) between these variables confirms the negative correlation between attenuation and the duration of the

GLOBAL

BREAST

ATTENUATION

Table I Correlation matrix of the frequency dependant attenuation with physiological and acquired parameters (study including 49 normals with the same age distribution) Attenuation

Attenuation

Duration of genital life

1.0

Dur. gen. life

Phase

Pers.

breast

Personal breast history

-0.27

-0.46

0.09

0.09

1.0

0.39

0.64

-0.08

-0.16

1.0

0.15

-0.003

-0.16

1.0

0.11

-0.16

1.0

0.02

Pregnancy breast hist.

Family breast history

-0.54

Phase

Fam.

Pregnancy

hist.

1.0

genital life (DGL). It also shows a negative attenuation and pregnancies. In fact, as one might Correlation and pregnancies are partly correlated. the other parameters is weaker or is not significant.

correlation between have expected, DGL of attenuation with

To complete this approach, we present the results of a multiple linear regression performed on the data. This regression is of a classic which expresses the logarithmic attenuation as a linear type combination of the previously-defined variables ( Eq. (2) ) : Log ml

= I + xi I ci vi I

where A represents the attenuation coefficient p multiplied by 100 (in order to compute integers), 1 is the intercept of the regression and c i are the

relative

regression

coefficients

of

each

variable

Vi

DGL,

(

pregnancies, phase, family breast history, personal breast history). Table II gives the ci coefficients together with the corresponding t-test and significance level. The DGL is the most significant subset that explains the attenuation variation ( p < 0.05). These parameters account for 37 % of the variation. This distribution

variation might be of mammary tissues

related versus

53

to the modification age. On one hand,

of

the we know

BERGER El’ AL

Multiple

I

Table II linear regression of Log (Attenuation x 100 ) : coefficients regression for each variable , t test, significance level

Variable

Duration

of

genital

Coefficient

t-test

-0.0104 -0.0962 -0.0361 0.0605 0.28E-3

-2.355 -1.045 0.953 0.724 0.002

life

Pregnancy Phase Family breast history Personal breast history

MULTIPLE

of the

P

0.023 0.31 0.35 0.47 0.99

R2 = 0.37

that at a certain period in the woman’s life, involuted ductal-alveolar structures are replaced by adipose tissue. On the other hand, there appears to be general agreement that attenuation is correlated with fatty changes (liver [17], muscle [18]). In this study also, the global attenuation of the breast could be partly influenced by fatty changes , vs. the duration of the genital life.

PRELIMINARY 1.

RESULTS

Benign

ON PATHOLOGICAL

CASES

breast disease

This preliminary study includes only 10 pathological cases. Those cases consist of benign breast disease. Our goal was not to differentiate among those cases themselves, but to differentiate normal from abnormal breasts. The 10 values are presented together with the attenuation values of the normals as a function of the DGL in figure 6. Each population is discriminated by a different parameter in the multiple linear regression., This parameter is called the type and will be equal to 0 for normal cases and to 1 for pathological cases. The regression is performed on the whole set of data including this additive parameter. Results are presented in table III. They show that the type is the most significant parameter that explains the attenuation variation : p < 0.01). All parameters together explain 36% of (significance level the variation. 2. Single large

We have fibrocystic

case of fibrocystic also investigated disease. The

disease a woman pathology

54

presenting affected

a particularly both breasts,

GLOBAL BREAST ATTENUATION

G Eg

3.2

-

2.8

-

2.4

Y s m

A

-

A

1.6

-

1.2

A

0.8

-

+?

0.4

-

A

AAA AAA

A A

9

AA

4 0

-

5E=

III

Af A AA&

2-

$ i: 2

A

normal volunteers benign breast diseases

A

AA

A q

0

20

10

30

A A

A

40

50

60

Duration of genital life : DGL (years) Fig. 6

Breast attenuation as a function of the duration of genital life for 10 benign breast diseases with respect to the attenuation distribution in a population of 49 normal volunteers.

particularly on the right side, where a large solid nodule of 5 The global attenuation of the right breast diameter was present. 0.38 dB/cm MHz, that of the left breast was 0.61 dB/cm MHz. nodule itself has very low attenuation, 0.02 dB/cm MHz. We note these values are all out of the normal range (Eq. (1)).

cm was The that

Table Ill linear regression of log (Attenuation x 100) : coefficients of the regression for each variable, t test, significance level. The parameter type discriminate the normals (49) from those with the benign breast diseases (1 0)

Multiple

Variable

Coefficient

Type Duration

of

genital

life

Pregnancy Phase Family breast history Personal breast history

1 MULTIPLE

55

P

-0.3889 -0.0088

-2.715 -1.97

0.009 0.054

-0.0934 -0.0331 0.0054

-1.080 -0.79 0.600

0.280 0.43 0.95

0.26

0.79

0.0342

R2 = 0.36

t-test

BERGER

ET

AL

CONCLUSION

There is a need for a quantitative noninvasive technique that would permit the characterization and the followup of benign breast diseases. We have investigated the attenuation coefficient. The investigation of a control group shows large inter-individual variations. A multiple linear regression explains some dependence on the duration of the genital life. This implies that any conclusions on pathological cases will require references to a large control group. We may therefore expect to determine normal breast attenuation limits for each subgroup, based on age, duration of genital life, pregnancies, etc..On the other hand, this observation has wider implications in the field of attenuation and tissue characterization. It clearly shows that the acoustic parameters can be modified by physiological and acquired parameters. This is similar to other clinical indices where reference to large control groups and large physiological variations are required. Nevertheless, in a preliminary study that included 10 benign breast diseases, it appears that the attenuation was partly correlated with the type of breast : normal or pathological. Observations performed on a breast with very large fibrocystic disease shows that the attenuation parameter quantifies the involvement and might be interesting for followup purposes.

RFsERENcEs [II

using Finette, S., Bleier, A., and Swindell, W. , Breast tissue classification diagnostic ultrasound and pattern recognition techniques : I methods of pattern recognition Ultrasonic Imaging 5 , 55 - 70 (1983).

[21

A., Swindell, Finette, S., Bleier, classification using diagnostic techniques. II Experimental results,

[31

Greenleaf, Experiment (Academic

[41

Measurement of ultrasonic attenuation Parker, K.J., and Waag, R.C. , within regions selected from B-Scans images, IEEE Trans. Biom. Eng. 30 , 431-437 (1983)

[51

Sehgal, C. M., and Greenleaf, J.F. , Derivation images from speed and attenuation reconstructions, 98 - 99 (Abstract) (1985).

[61

Berger, G., measurements

W., and ultrasound Ultrasonic

Harber, K. , Breast tissue and pattern recognition Imaging 5 , 71 - 86 (1983)

Computerized Transmission Tomography, in Methods of J. F., Physics Ultrasound, P.D. Edmonds, ed., Vol 19, pp. 563 - 589 Press, New York, 1981).

Perrin, for

J., and Bourgoin D. Tissue Characterization

56

, .

of

scatter and absorption Ultrasonic Imaging 7 ,

Attenuation principles and Tissue in Ultrasonic

GLOBAL BREAST ATTENUATION

Characterisation, Nijhoff Publishers,

J.M. Thijssen, and D. Nicholas The netherlands, 1982)

eds., pp. 6-18

( Martinus

171

DuPont W.D., and Page D.L., Breast cancer risk associated with proliferative disease, age at first birth, and a family history of breast cancer, American J. Epidemiology, 125, 769-779 (1987)

[81

Brinton L.A., hormonal risk (1982)

[91

Laugier, P., Berger, G., Fink, M., and Perrin, J. , Specular reflector noise effect and correction for in vivo attenuation estimation, Ultrasonic Imaging 7 , 277 - 292 (1985).

[ 10 ]

Berger, G., Laugier, P., Fink, M., and Perrin, J. , Optimal ultrasound attenuation estimation and application to the Duchenne Muscular Dystrophy Carriers, Ultrasonic Imaging (1987).

precision in detection of 9 , 1 - 17

[ 11 ]

Laugier, focused Imaging

correction for U 1r ras o ni c

[ 12 ]

signal processing for in Fink, M., Hottier, F. and Cardoso, J.F. , Ultrasonic vivo attenuation measurements : short time Fourier analysis, Ultrasonic Imaging 5 , 117-135 (1983).

[ 13 ]

Hottier, F., and Bematets, J. L. biological tissues, Acta Electronica

[ 14 ]

Cardoso, J.F. and Fink, M., Diffraction correction in pulse echo attenuation measurements, in 1983 IEEE Ultrasonics Symposium Proceedings ( pp. 841-846( IEEE Cat No 83 CH 1941-l,IEEE, New York, 1983).

[ 15 ]

Cloostermans, M.J.T.M., and Thijssen, J.M. , A beam corrected estimation of the frequency dependant attenuation of biological tissues from backscattered ultrasound, Ultrasonic Imaging 5 , 136 - 174 (1983)

[ 16 ]

Kuc, R., Bounds on estimating regions from reflected ultrasound,

[ 17 ]

Taylor, K.J., Riely, C.A., Hammers, L., Flax, S., Weltin, G., Garsia-Tsao, G., Corm, H.O., Kuc, R., and Barwick, K.B., Quantitative US attenuation in normal liver and patients with diffuse liver disease : importance of fat, Radiology 160, 65-75 (1986)

[ 18 ]

Berger, G., Laugier, P., Leroy, A. , Fink, M., Roucayrol, J.C., and Perrin, J. , Correlation between ultrasound attenuation in muscle and pathological fatty infiltration, Ultrasonic Imaging 9, 66 (1987).

Hoover R., and Fraumeni J.F., Interaction of familial and factors for braest cancer, J. Nutl. Cancer Inst., 69, 817-822

P., Berger, transducers 9, 248-259

G., Fink, M., and Perrin, J. , Diffraction in attenuation measurements in vivo, (1987)

, Estimation 26, 33-58

of ultrasonic (1984).

the acoustic attenuation Proc. IEEE 73, 1159-1168

57

attenuation

of small (1985)

:

in

tissue