- Email: [email protected]
Performance testing of medical ultrasound equipment: fundamental vs. harmonic mode
Ultrasonics 40 (2002) 585–591 www.elsevier.com/locate/ultras
Performance testing of medical ultrasound equipment: fundamental vs. harmonic mode M.C. van Wijk, J.M. Thijssen
*
435 Clinical Physics Laboratory, University Children’s Hospital and Heart Center, P.O. Box 9101, 6500 HB Nijmegen, Netherlands
Abstract Assessment of the performance of medical ultrasound equipment is generally based on the image quality in fundamental mode. Recent development of the so-called tissue harmonic imaging (THI) mode induces the need for assessment of differences in the quality of imaging in THI vs. fundamental imaging mode. Quality features to be tested are sensitivity (penetration depth), spatial resolution, contrast resolution, lesion signal-to-noise ratio, and tissue-to-clutter ratio (TCR). These features are explained and examples are shown. The main conclusion from a comparison of the results for the two imaging modes might be that when using THI improvement of TCR, in particular in the near field, is obtained at the expense of a loss in axial resolution. Furthermore, lesion detection is not significantly improved. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Ultrasound; Medical imaging; Performance testing; Harmonic imaging; Clutter; Lesion detection
1. Introduction Quality assessment (QA) as part of a quality assurance protocol has become of primary interest in hospitals, at least in those where (ISO) certification has been introduced. It might be useful to distinguish two levels of QA: (1) testing of technical specifications of the equipment, and (2) testing of the performance from the user’s point of view. In case of medical ultrasound, the latter may be summarized by ‘‘imaging performance’’. A final aspect of QA procedures and protocols that have been published so far [1–5] is the subjective nature of the testing procedures. For this reason, the testing results become strongly dependent on the chosen settings of the equipment and on the visual interpretation of displayed information during the testing, i.e. the observer is part of the procedure. More recently, the authors have developed a protocol [6] based on QA by software algorithms applied to stored digital ultrasound images, while using reproducible equipment settings. The aim of this contribution is to apply these methods in a comparison of fundamental
*
Corresponding author. Tel.: +31-24-3619061; fax: +31-243616824. E-mail address: [email protected] (J.M. Thijssen).
and ‘‘tissue harmonic imaging’’ (THI) modes. Performance features that might be expected to depend on a choice of either of these modes are: contrast resolution, spatial resolution and clutter reduction. The concept of contrast resolution can be extended in various ways: (1) contrast-to-noise ratio (CNR) which relates contrast resolution to pixel gray level statistics (standard deviation), (2) lesion signal-to-noise ratio [7–9] (SNRL ) which quantifies the detectability of a lesion of a certain contrast and size. Clutter reduction was investigated by estimating the ‘‘tissue-to-clutter ratio’’ (TCR) of imaged voids (‘‘cysts’’) contained in a scattering tissue mimicking phantom. All these concepts are applied in this paper and a comparison between fundamental and harmonic performances is made.
2. Backgrounds 2.1. General Harmonic imaging has become possible due to the advent of wide band transducers. Some manufacturers have fully used the enhanced bandwidth to improve the image quality in fundamental mode (see Section 2.5). Other manufacturers introduced the software option to choose the transmit/receive frequency freely within the
0041-624X/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 1 - 6 2 4 X ( 0 2 ) 0 0 1 7 7 - 4
586
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
wide band, but then in narrow band operation. This has a disadvantage of less axial resolution in any of the frequencies chosen in fundamental mode. It is therefore necessary to consider both these ways of using wide band transducers when comparing fundamental and harmonic imaging modes. A further aspect is the influence of this choice on the sensitivity: because of the linear frequency dependence of the attenuation coefficient, transmission at a lower frequency would mean less attenuation. The authors have circumvented this problem to a certain extent by keeping the MI of the transmitted ultrasound pulse identical in both imaging modes. MI stand for ‘‘mechanical index’’ [10] and is defined as the maximum rarefaction pressure determined at the position of the maximum value of the derated pulse intensity integral. 2.2. Equipment settings Before starting QA measurements, several controls of the equipment have to be set in a fixed setting, with either adequate read-out, or reproducible marking of the control switch. These controls are: output, generally visible by means of the MI value (e.g. MI is set to 1.0). Gain (overall), preferably specified in dB at the display. Time gain compensation (TGC), this is a difficult control because no (useful) read-out is available in general. In some equipment the TGC is preset to a setting which compensates for an average attenuation in tissue. In this case the TGC slide rulers can be set in neutral (middle) position. In other equipment a reproducible setting might be a linear increase of sliding potentiometer positions with depth. Reject, i.e. a threshold, should be set in the (fixed) most useful position, it might be advisable to use the lowest threshold position. The post-processing options, e.g. contrast range, should be fixed. Contrast (dynamic) range is mostly specified on screen in dB, the default setting of the equipment can be used. The display option specifying the look-up table (LUT) for read-out of image memory should be set to a linear curve. This means that the logarithmic compression, involved in the preprocessing prior to AD-conversion, is being maintained. In this way, the relation of gray levels to echo levels in dB can be determined. Finally, the (single!) transmit focus should be set at the depth of the elevation focus. This elevation focus is estimated by using a ‘‘slice thickness’’ phantom [6].
disc) and plotting these against its known nominal contrast value in dB, a linear regression can de made. The slope of this line yields the ‘‘gamma’’ of the systems in gray levels per dB and the dB range corresponding to gray levels from 0 to 255 yields the contrast dynamic range. The definitions used are Contrast: C¼
jlB lL j ðlB þ lL Þ=2
ð1Þ
where lB and lL are mean gray levels in background and in lesions, respectively. Contrast-to-noise ratio: jlB lL j ffi CNR ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðr2B þ r2L Þ=2
ð2Þ
where r2B and r2L are variance of gray levels within background and lesion, respectively. SNRL (or: Mahalanobis distance), as defined below, was calculated after conversion of the gray levels to dB and subsequent inversion to linear echo levels (by taking the antilog of the dB levels): jhmB i hmL ij SNRL ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r2mB þ r2mL
ð3Þ
where hmB i, hmL i are the ensemble average of lesion and background echo levels, respectively. r2mB and r2mL are the variances of the average echo level of B and L, respectively. It can be shown [7], that SNRL contains the contrast, the CNR and the square root of the number of speckles within the lesion area. The involvement of the speckle size in SNRL means that the spatial resolution is involved in the detectability of a lesion. The contrast sensitivity assessment by computer software is sometimes called the ‘‘computational observer’’ analysis [11]. 2.4. Tissue-to-clutter ratio The TCR is estimated by measuring the mean gray level within liquid filled cylinders (voids, or cysts) within the phantom. This level is compared to the mean background level measured at the same depth and over the same circular area. Definition of TCR: hl i TCR ¼ 20 log10 B ð4Þ hlL i where h i means ensemble averaging, lB , lL are mean gray level of background and void, respectively.
2.3. Contrast resolution 2.5. Spatial resolution Some tissue mimicking test objects (e.g. ATS Labs, Bridgeport, CT) contain cylinders with fixed diameter and scattering levels different from the surrounding material (‘‘background’’). By measuring the mean gray level of each of the cylinders (in cross-section visible as a
This QA feature is measured by using the image of a thin wire in the focus. The full width at half maximum (FWHM) can be estimated in axial and in lateral direction by using the gamma (Section 2.3) determined
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
with the contrast measurements. The software developed by the authors enables a simple measurement by interpolated gray level profiles in dB [6]. The effects of harmonic vs. fundamental imaging on spatial resolution can be predicted by using the formulas valid for the FWHM in both directions [12]. The axial resolution follows from the FWHMax (in mm): FWHMax ¼ 0:66=Df
The resolution differences between fundamental and harmonic modes are therefore depending both on bandwidth and on central frequency. 3. Methods 3.1. Equipment
ð5Þ
where Df is 6 dB bandwidth (transmit/receive) in MHz. The lateral resolution is specified by the FWHMlat : FWHMlat ¼ 1:02cF =fc D
587
ð6Þ
where c is the speed of sound [mm/ls], F the focal depth [mm], fc the (nominal) central frequency (transmit/ receive) [MHz], and D the diameter of transducer [mm].
Equipment 1 is a high-end system, for general echographic work: System 5000 HDI (Philips/ATL, Bothell, USA) with a 4–2 MHz phased array transducer. The application mode ‘‘general’’ was installed for the measurements. The transducer was operated at 4–2 MHz in fundamental mode and in harmonic mode at 4 MHz receive frequency. The output was set at MI ¼ 1:0. Equipment 2 used in this paper was a high-end scanner for cardiovascular applications: System Five
Fig. 1. Top: contrast sensitivity curves for equipment 1 (Philips/ATL) obtained in fundamental and in harmonic imaging modes, respectively. Bottom: same for equipment 2 (General Electric/Vingmed).
588
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
(General Electric/Vingmed Ultrasound Europe, Solingen, Germany) with a phased array transducer 3.5 MHz. The equipment was used in ‘‘Pediatric’’ mode. The transducer was operated in fundamental mode at 4 MHz and also in harmonic mode at 4 MHz, receive frequency. MI was set at 1.0. 3.2. Tissue mimicking phantom The phantom used in the QAs was a general purpose imaging phantom #539 (ATS Laboratories Inc., Bridgeport, CT, USA). This phantom consists of a urethane base material with scattering microstructures. It contains several objects for quality measurements, like thin wires, liquid filled cylinders and larger diameter cylinders with scattering strength somewhat higher and somewhat lower than the surrounding base material.
Table 1 Contrast-to-noise ratio Value [dB]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
15.00 6.00 3.00 3.00 6.00 15.00
2.79 1.21 0.87 0.38 1.17 2.41
2.31 1.07 0.72 0.74 1.35 2.74
2.84 1.80 0.65 1.14 1.52 4.00
1.46 1.05 0.47 0.97 1.17 2.95
produced a little bit ‘‘harder’’ image (more black/white) than the harmonic mode. Overall, it can be concluded that the harmonic mode of ultrasound equipment does not change the contrast sensitivity dramatically. 4.2. Contrast-to-noise ratio
4. Results 4.1. Contrast sensitivity For this measurement five cross-sectional images of each cylinder were acquired and analysed with equal size regions of interest (ROIs). The measurements obtained from each object were averaged and the standard deviation calculated. As can be seen from Fig. 1 (top), equipment 1 yields a good linear contrast curve and the slope of the regression line remains nearly constant when switching to harmonic mode. Some bias can be remarked of about 20 gray levels. In Fig. 1, bottom, the results for equipment 2 display only a slight change of the slope. However, this change of gamma gave a little perceptual change of the image. The fundamental mode
The data of the contrast sensitivity measurement were used next to estimate the CNR (cf., Eq. (2)). In Fig. 2, the results obtained for equipment 1 are shown. The results for fundamental and harmonic modes are practically the same. The results obtained for both systems are summarized in Table 1. 4.3. Lesion signal-to-noise ratio From the same measurements as the contrast sensitivity, the SNRL was calculated (cf., Eq. (3)). The results for equipment 1 are shown in Fig. 3 and in Table 2. From this figure, it can be seen that SNRL is practically linearly dependent on the echo level. However, a systematic difference between the fundamental and harmonic modes seems to be absent.
Fig. 2. CNR vs. nominal echo level (dB) for fundamental and harmonic modes, respectively (equipment 1).
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
589
Fig. 3. SNRL (Mahalanobis distance) vs. nominal echo level of discs for fundamental and harmonic modes, respectively (equipment 1).
Table 2 Signal-to-noise ratio SNRL Value [dB]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
15.00 6.00 3.00 3.00 6.00 15.00
14.25 3.91 2.91 1.79 4.78 7.97
10.23 5.60 3.75 3.00 4.90 7.18
31.72 10.65 5.33 13.09 16.59 27.39
9.60 7.12 2.94 4.57 5.29 7.93
On each depth and with the same ROI size, the mean pixel value was measured inside and outside the void. The TCR was calculated in dB, as defined in Eq. (4). Fig. 4 (cf., also Table 3) shows that an advantage of equipment 1 is a perceptual improvement of the imaged voids in harmonic mode, in particular in the near field. This corresponds to an improvement of the TCR by about 4 dB at a depth of 20 mm. Considering the results of equipment 2 (Table 3), an overall improvement of the TCR can be denoted, on average by plus 3 dB.
4.4. Tissue-to-clutter ratio
4.5. Spatial resolution
For this measurement, three non-echoic objects (or voids) of 7 mm were scanned. The depths are respectively 20, 40 and 60 mm. The azimuth focus was maintained at 40 mm, which coincides with the second void.
Comparing the spatial resolution of the investigated ultrasound machines in fundamental mode, they appeared to be almost identical (Fig. 5). Switching to harmonic mode evoked some remarkable differences.
Fig. 4. TCR for cystic objects at depths 20–60 mm, equipment 1.
590
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
The axial FWHM (Eq. (5)) on the contrary has been more than doubled, which might be considered as one of the drawbacks of harmonic imaging.
Table 3 Tissue-to-clutter ratio TCR [dB] Depth [mm]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
20.00 40.00 60.00
13.43 14.04 11.56
17.43 13.99 11.69
6.28 9.81 12.43
10.56 13.34 14.47
Investigation of the results of the equipment 1, a decrease of the FWHM in the lateral direction is noticeable, which is higher than for equipment 2. The axial resolution however, is slightly worse than in the fundamental mode but still much better than equipment 2. Considering equipment 2 in harmonic mode, it provided twice the lateral resolution as compared to the fundamental mode, due to the increased effective aperture (Eq. (6)) halving the lateral FWHM size of the PSF.
5. Discussion and conclusions Considering the results, some conclusions might be drawn about the fundamental and the harmonic modes of these machines. It is well known that, when switching the equipment to harmonic mode, some trade-off must exist between the lateral and the axial resolution. Generally speaking, the lateral resolution improves due to increased receive frequency (Eq. (6)) and the axial resolution degrades due to reduced receive bandwidth (Eq. (5)). For equipment 1, the advantage of the harmonic mode is an improvement of the TCR ratio in the near field (Fig. 4) and no loss of the detectability of the
Fig. 5. Top: axial (dark bars) and lateral (light bars) resolution obtained in fundamental mode (left) and harmonic mode (right), equipment 1. Bottom: same for equipment 2.
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
lesions. For equipment 2, the detectability appeared much lower in harmonic mode than in fundamental mode. This result is in line with the visual impression of the images. It is, therefore, important to realise, what type of ultrasound equipment is wanted for a certain application area. The reason why the SNRL of equipment 1 remains constant to a first approximation in harmonic vs. fundamental mode is mainly due to the fact that the twodimensional size (area) of the PSF remains constant (cf., Fig. 5). This means that the number of speckles within the lesion area remains constant. The authors want to stress that the results discussed in this paper for the two echographic systems should not be considered as a ‘‘consumers’’ report but merely as representative examples of performance testing.
References [1] The AIUM 100 mm test object and recommended procedures for its use, American Institute of Ultrasound in Medicine (AIUM), Rockville, 1974. [2] C.R. Hill (Ed.), Methods of measuring the performance of ultrasonic pulse-echo diagnostic equipment (Discussion document of International Electrotechnical Committee), Ultrasound Med. Biol. 2 (1977) 343–350.
591
[3] P.L. Carson, J.A. Zagzebski, Pulse-Echo Ultrasound Imaging Systems: Performance Tests and Criteria, Am. Inst. Physics, New York, 1977. [4] M.M. Goodsitt, P.L. Carson, S. Witt, D.L. Hykes, J.M. Kofler, Real-time B-mode ultrasound quality control test procedures, Med. Phys. 25 (1998) 1385–1406. [5] N.J. Dudley, K. Griffith, G. Houldsworth, M. Holloway, M.A. Dunn, A review of two alternative ultrasound quality assurance programs, Eur. J. Ultrasound 12 (2001) 233–245. [6] J.M. Thijssen, M.C. van Wijk, M.H. Cuypers, Performance testing of medical echo/Doppler equipment, Eur J. Ultrasound, accepted. [7] S.W. Smith, R.F. Wagner, J.M. Sandrik, H. Lopez, Low contrast detectability and contrast/detail analysis in medical ultrasound, IEEE Trans. Sonics Ultrason. 30 (1983) 164–173. [8] J.M. Thijssen, B.J. Oosterveld, R.F. Wagner, Gray level transforms and lesion detectability in echographic images, Ultrasonic Imag. 10 (1988) 171–195. [9] R.F. Wagner, M.F. Insana, D.G. Brown, Unified approach to the detection and classification of speckle texture in diagnostic ultrasound, Opt. Eng. 25 (1986) 738–742. [10] American Institute of Ultrasound in Medicine, Standard for RealTime Display of Thermal and Mechanical Acoustic output Indices on Diagnostic Ultrasound Equipment, AIUM/NEMA, Rockville/ Washington, 1992. [11] H. Lopez, M.H. Loew, D.J. Goodenough, Objective analysis of ultrasound images by use of a computational observer, IEEE Trans. Med. Imag. 11 (1992) 496–506. [12] J.M. Thijssen, B.J. Oosterveld, Performance of echographic equipment and potentials for tissue characterization, in: M.A. Viergever, R. Todd-Prokopek (Eds.), Computer Science in Medical Imaging, Springer, Berlin, 1988, pp. 455–468.
Performance testing of medical ultrasound equipment: fundamental vs. harmonic mode M.C. van Wijk, J.M. Thijssen
*
435 Clinical Physics Laboratory, University Children’s Hospital and Heart Center, P.O. Box 9101, 6500 HB Nijmegen, Netherlands
Abstract Assessment of the performance of medical ultrasound equipment is generally based on the image quality in fundamental mode. Recent development of the so-called tissue harmonic imaging (THI) mode induces the need for assessment of differences in the quality of imaging in THI vs. fundamental imaging mode. Quality features to be tested are sensitivity (penetration depth), spatial resolution, contrast resolution, lesion signal-to-noise ratio, and tissue-to-clutter ratio (TCR). These features are explained and examples are shown. The main conclusion from a comparison of the results for the two imaging modes might be that when using THI improvement of TCR, in particular in the near field, is obtained at the expense of a loss in axial resolution. Furthermore, lesion detection is not significantly improved. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Ultrasound; Medical imaging; Performance testing; Harmonic imaging; Clutter; Lesion detection
1. Introduction Quality assessment (QA) as part of a quality assurance protocol has become of primary interest in hospitals, at least in those where (ISO) certification has been introduced. It might be useful to distinguish two levels of QA: (1) testing of technical specifications of the equipment, and (2) testing of the performance from the user’s point of view. In case of medical ultrasound, the latter may be summarized by ‘‘imaging performance’’. A final aspect of QA procedures and protocols that have been published so far [1–5] is the subjective nature of the testing procedures. For this reason, the testing results become strongly dependent on the chosen settings of the equipment and on the visual interpretation of displayed information during the testing, i.e. the observer is part of the procedure. More recently, the authors have developed a protocol [6] based on QA by software algorithms applied to stored digital ultrasound images, while using reproducible equipment settings. The aim of this contribution is to apply these methods in a comparison of fundamental
*
Corresponding author. Tel.: +31-24-3619061; fax: +31-243616824. E-mail address: [email protected] (J.M. Thijssen).
and ‘‘tissue harmonic imaging’’ (THI) modes. Performance features that might be expected to depend on a choice of either of these modes are: contrast resolution, spatial resolution and clutter reduction. The concept of contrast resolution can be extended in various ways: (1) contrast-to-noise ratio (CNR) which relates contrast resolution to pixel gray level statistics (standard deviation), (2) lesion signal-to-noise ratio [7–9] (SNRL ) which quantifies the detectability of a lesion of a certain contrast and size. Clutter reduction was investigated by estimating the ‘‘tissue-to-clutter ratio’’ (TCR) of imaged voids (‘‘cysts’’) contained in a scattering tissue mimicking phantom. All these concepts are applied in this paper and a comparison between fundamental and harmonic performances is made.
2. Backgrounds 2.1. General Harmonic imaging has become possible due to the advent of wide band transducers. Some manufacturers have fully used the enhanced bandwidth to improve the image quality in fundamental mode (see Section 2.5). Other manufacturers introduced the software option to choose the transmit/receive frequency freely within the
0041-624X/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 1 - 6 2 4 X ( 0 2 ) 0 0 1 7 7 - 4
586
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
wide band, but then in narrow band operation. This has a disadvantage of less axial resolution in any of the frequencies chosen in fundamental mode. It is therefore necessary to consider both these ways of using wide band transducers when comparing fundamental and harmonic imaging modes. A further aspect is the influence of this choice on the sensitivity: because of the linear frequency dependence of the attenuation coefficient, transmission at a lower frequency would mean less attenuation. The authors have circumvented this problem to a certain extent by keeping the MI of the transmitted ultrasound pulse identical in both imaging modes. MI stand for ‘‘mechanical index’’ [10] and is defined as the maximum rarefaction pressure determined at the position of the maximum value of the derated pulse intensity integral. 2.2. Equipment settings Before starting QA measurements, several controls of the equipment have to be set in a fixed setting, with either adequate read-out, or reproducible marking of the control switch. These controls are: output, generally visible by means of the MI value (e.g. MI is set to 1.0). Gain (overall), preferably specified in dB at the display. Time gain compensation (TGC), this is a difficult control because no (useful) read-out is available in general. In some equipment the TGC is preset to a setting which compensates for an average attenuation in tissue. In this case the TGC slide rulers can be set in neutral (middle) position. In other equipment a reproducible setting might be a linear increase of sliding potentiometer positions with depth. Reject, i.e. a threshold, should be set in the (fixed) most useful position, it might be advisable to use the lowest threshold position. The post-processing options, e.g. contrast range, should be fixed. Contrast (dynamic) range is mostly specified on screen in dB, the default setting of the equipment can be used. The display option specifying the look-up table (LUT) for read-out of image memory should be set to a linear curve. This means that the logarithmic compression, involved in the preprocessing prior to AD-conversion, is being maintained. In this way, the relation of gray levels to echo levels in dB can be determined. Finally, the (single!) transmit focus should be set at the depth of the elevation focus. This elevation focus is estimated by using a ‘‘slice thickness’’ phantom [6].
disc) and plotting these against its known nominal contrast value in dB, a linear regression can de made. The slope of this line yields the ‘‘gamma’’ of the systems in gray levels per dB and the dB range corresponding to gray levels from 0 to 255 yields the contrast dynamic range. The definitions used are Contrast: C¼
jlB lL j ðlB þ lL Þ=2
ð1Þ
where lB and lL are mean gray levels in background and in lesions, respectively. Contrast-to-noise ratio: jlB lL j ffi CNR ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðr2B þ r2L Þ=2
ð2Þ
where r2B and r2L are variance of gray levels within background and lesion, respectively. SNRL (or: Mahalanobis distance), as defined below, was calculated after conversion of the gray levels to dB and subsequent inversion to linear echo levels (by taking the antilog of the dB levels): jhmB i hmL ij SNRL ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r2mB þ r2mL
ð3Þ
where hmB i, hmL i are the ensemble average of lesion and background echo levels, respectively. r2mB and r2mL are the variances of the average echo level of B and L, respectively. It can be shown [7], that SNRL contains the contrast, the CNR and the square root of the number of speckles within the lesion area. The involvement of the speckle size in SNRL means that the spatial resolution is involved in the detectability of a lesion. The contrast sensitivity assessment by computer software is sometimes called the ‘‘computational observer’’ analysis [11]. 2.4. Tissue-to-clutter ratio The TCR is estimated by measuring the mean gray level within liquid filled cylinders (voids, or cysts) within the phantom. This level is compared to the mean background level measured at the same depth and over the same circular area. Definition of TCR: hl i TCR ¼ 20 log10 B ð4Þ hlL i where h i means ensemble averaging, lB , lL are mean gray level of background and void, respectively.
2.3. Contrast resolution 2.5. Spatial resolution Some tissue mimicking test objects (e.g. ATS Labs, Bridgeport, CT) contain cylinders with fixed diameter and scattering levels different from the surrounding material (‘‘background’’). By measuring the mean gray level of each of the cylinders (in cross-section visible as a
This QA feature is measured by using the image of a thin wire in the focus. The full width at half maximum (FWHM) can be estimated in axial and in lateral direction by using the gamma (Section 2.3) determined
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
with the contrast measurements. The software developed by the authors enables a simple measurement by interpolated gray level profiles in dB [6]. The effects of harmonic vs. fundamental imaging on spatial resolution can be predicted by using the formulas valid for the FWHM in both directions [12]. The axial resolution follows from the FWHMax (in mm): FWHMax ¼ 0:66=Df
The resolution differences between fundamental and harmonic modes are therefore depending both on bandwidth and on central frequency. 3. Methods 3.1. Equipment
ð5Þ
where Df is 6 dB bandwidth (transmit/receive) in MHz. The lateral resolution is specified by the FWHMlat : FWHMlat ¼ 1:02cF =fc D
587
ð6Þ
where c is the speed of sound [mm/ls], F the focal depth [mm], fc the (nominal) central frequency (transmit/ receive) [MHz], and D the diameter of transducer [mm].
Equipment 1 is a high-end system, for general echographic work: System 5000 HDI (Philips/ATL, Bothell, USA) with a 4–2 MHz phased array transducer. The application mode ‘‘general’’ was installed for the measurements. The transducer was operated at 4–2 MHz in fundamental mode and in harmonic mode at 4 MHz receive frequency. The output was set at MI ¼ 1:0. Equipment 2 used in this paper was a high-end scanner for cardiovascular applications: System Five
Fig. 1. Top: contrast sensitivity curves for equipment 1 (Philips/ATL) obtained in fundamental and in harmonic imaging modes, respectively. Bottom: same for equipment 2 (General Electric/Vingmed).
588
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
(General Electric/Vingmed Ultrasound Europe, Solingen, Germany) with a phased array transducer 3.5 MHz. The equipment was used in ‘‘Pediatric’’ mode. The transducer was operated in fundamental mode at 4 MHz and also in harmonic mode at 4 MHz, receive frequency. MI was set at 1.0. 3.2. Tissue mimicking phantom The phantom used in the QAs was a general purpose imaging phantom #539 (ATS Laboratories Inc., Bridgeport, CT, USA). This phantom consists of a urethane base material with scattering microstructures. It contains several objects for quality measurements, like thin wires, liquid filled cylinders and larger diameter cylinders with scattering strength somewhat higher and somewhat lower than the surrounding base material.
Table 1 Contrast-to-noise ratio Value [dB]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
15.00 6.00 3.00 3.00 6.00 15.00
2.79 1.21 0.87 0.38 1.17 2.41
2.31 1.07 0.72 0.74 1.35 2.74
2.84 1.80 0.65 1.14 1.52 4.00
1.46 1.05 0.47 0.97 1.17 2.95
produced a little bit ‘‘harder’’ image (more black/white) than the harmonic mode. Overall, it can be concluded that the harmonic mode of ultrasound equipment does not change the contrast sensitivity dramatically. 4.2. Contrast-to-noise ratio
4. Results 4.1. Contrast sensitivity For this measurement five cross-sectional images of each cylinder were acquired and analysed with equal size regions of interest (ROIs). The measurements obtained from each object were averaged and the standard deviation calculated. As can be seen from Fig. 1 (top), equipment 1 yields a good linear contrast curve and the slope of the regression line remains nearly constant when switching to harmonic mode. Some bias can be remarked of about 20 gray levels. In Fig. 1, bottom, the results for equipment 2 display only a slight change of the slope. However, this change of gamma gave a little perceptual change of the image. The fundamental mode
The data of the contrast sensitivity measurement were used next to estimate the CNR (cf., Eq. (2)). In Fig. 2, the results obtained for equipment 1 are shown. The results for fundamental and harmonic modes are practically the same. The results obtained for both systems are summarized in Table 1. 4.3. Lesion signal-to-noise ratio From the same measurements as the contrast sensitivity, the SNRL was calculated (cf., Eq. (3)). The results for equipment 1 are shown in Fig. 3 and in Table 2. From this figure, it can be seen that SNRL is practically linearly dependent on the echo level. However, a systematic difference between the fundamental and harmonic modes seems to be absent.
Fig. 2. CNR vs. nominal echo level (dB) for fundamental and harmonic modes, respectively (equipment 1).
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
589
Fig. 3. SNRL (Mahalanobis distance) vs. nominal echo level of discs for fundamental and harmonic modes, respectively (equipment 1).
Table 2 Signal-to-noise ratio SNRL Value [dB]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
15.00 6.00 3.00 3.00 6.00 15.00
14.25 3.91 2.91 1.79 4.78 7.97
10.23 5.60 3.75 3.00 4.90 7.18
31.72 10.65 5.33 13.09 16.59 27.39
9.60 7.12 2.94 4.57 5.29 7.93
On each depth and with the same ROI size, the mean pixel value was measured inside and outside the void. The TCR was calculated in dB, as defined in Eq. (4). Fig. 4 (cf., also Table 3) shows that an advantage of equipment 1 is a perceptual improvement of the imaged voids in harmonic mode, in particular in the near field. This corresponds to an improvement of the TCR by about 4 dB at a depth of 20 mm. Considering the results of equipment 2 (Table 3), an overall improvement of the TCR can be denoted, on average by plus 3 dB.
4.4. Tissue-to-clutter ratio
4.5. Spatial resolution
For this measurement, three non-echoic objects (or voids) of 7 mm were scanned. The depths are respectively 20, 40 and 60 mm. The azimuth focus was maintained at 40 mm, which coincides with the second void.
Comparing the spatial resolution of the investigated ultrasound machines in fundamental mode, they appeared to be almost identical (Fig. 5). Switching to harmonic mode evoked some remarkable differences.
Fig. 4. TCR for cystic objects at depths 20–60 mm, equipment 1.
590
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
The axial FWHM (Eq. (5)) on the contrary has been more than doubled, which might be considered as one of the drawbacks of harmonic imaging.
Table 3 Tissue-to-clutter ratio TCR [dB] Depth [mm]
Equipment 1
Equipment 2
Fundamental
Harmonic
Fundamental
Harmonic
20.00 40.00 60.00
13.43 14.04 11.56
17.43 13.99 11.69
6.28 9.81 12.43
10.56 13.34 14.47
Investigation of the results of the equipment 1, a decrease of the FWHM in the lateral direction is noticeable, which is higher than for equipment 2. The axial resolution however, is slightly worse than in the fundamental mode but still much better than equipment 2. Considering equipment 2 in harmonic mode, it provided twice the lateral resolution as compared to the fundamental mode, due to the increased effective aperture (Eq. (6)) halving the lateral FWHM size of the PSF.
5. Discussion and conclusions Considering the results, some conclusions might be drawn about the fundamental and the harmonic modes of these machines. It is well known that, when switching the equipment to harmonic mode, some trade-off must exist between the lateral and the axial resolution. Generally speaking, the lateral resolution improves due to increased receive frequency (Eq. (6)) and the axial resolution degrades due to reduced receive bandwidth (Eq. (5)). For equipment 1, the advantage of the harmonic mode is an improvement of the TCR ratio in the near field (Fig. 4) and no loss of the detectability of the
Fig. 5. Top: axial (dark bars) and lateral (light bars) resolution obtained in fundamental mode (left) and harmonic mode (right), equipment 1. Bottom: same for equipment 2.
M.C. van Wijk, J.M. Thijssen / Ultrasonics 40 (2002) 585–591
lesions. For equipment 2, the detectability appeared much lower in harmonic mode than in fundamental mode. This result is in line with the visual impression of the images. It is, therefore, important to realise, what type of ultrasound equipment is wanted for a certain application area. The reason why the SNRL of equipment 1 remains constant to a first approximation in harmonic vs. fundamental mode is mainly due to the fact that the twodimensional size (area) of the PSF remains constant (cf., Fig. 5). This means that the number of speckles within the lesion area remains constant. The authors want to stress that the results discussed in this paper for the two echographic systems should not be considered as a ‘‘consumers’’ report but merely as representative examples of performance testing.
References [1] The AIUM 100 mm test object and recommended procedures for its use, American Institute of Ultrasound in Medicine (AIUM), Rockville, 1974. [2] C.R. Hill (Ed.), Methods of measuring the performance of ultrasonic pulse-echo diagnostic equipment (Discussion document of International Electrotechnical Committee), Ultrasound Med. Biol. 2 (1977) 343–350.
591
[3] P.L. Carson, J.A. Zagzebski, Pulse-Echo Ultrasound Imaging Systems: Performance Tests and Criteria, Am. Inst. Physics, New York, 1977. [4] M.M. Goodsitt, P.L. Carson, S. Witt, D.L. Hykes, J.M. Kofler, Real-time B-mode ultrasound quality control test procedures, Med. Phys. 25 (1998) 1385–1406. [5] N.J. Dudley, K. Griffith, G. Houldsworth, M. Holloway, M.A. Dunn, A review of two alternative ultrasound quality assurance programs, Eur. J. Ultrasound 12 (2001) 233–245. [6] J.M. Thijssen, M.C. van Wijk, M.H. Cuypers, Performance testing of medical echo/Doppler equipment, Eur J. Ultrasound, accepted. [7] S.W. Smith, R.F. Wagner, J.M. Sandrik, H. Lopez, Low contrast detectability and contrast/detail analysis in medical ultrasound, IEEE Trans. Sonics Ultrason. 30 (1983) 164–173. [8] J.M. Thijssen, B.J. Oosterveld, R.F. Wagner, Gray level transforms and lesion detectability in echographic images, Ultrasonic Imag. 10 (1988) 171–195. [9] R.F. Wagner, M.F. Insana, D.G. Brown, Unified approach to the detection and classification of speckle texture in diagnostic ultrasound, Opt. Eng. 25 (1986) 738–742. [10] American Institute of Ultrasound in Medicine, Standard for RealTime Display of Thermal and Mechanical Acoustic output Indices on Diagnostic Ultrasound Equipment, AIUM/NEMA, Rockville/ Washington, 1992. [11] H. Lopez, M.H. Loew, D.J. Goodenough, Objective analysis of ultrasound images by use of a computational observer, IEEE Trans. Med. Imag. 11 (1992) 496–506. [12] J.M. Thijssen, B.J. Oosterveld, Performance of echographic equipment and potentials for tissue characterization, in: M.A. Viergever, R. Todd-Prokopek (Eds.), Computer Science in Medical Imaging, Springer, Berlin, 1988, pp. 455–468.