Investigation of gain optimization technique in Doppler ultrasound system

Transcription

1 PAPER #2009 The Acoustical Society of Jaan Investigation of gain otimization technique in Doler ultrasound system Tatsuro Baba Toshiba Medical Systems Cororation ( Received 14 February 2008, Acceted for ublication 13 May 2008 ) Abstract: In this aer the author describes a high sensitivity Doler ultrasound system based on the rincile of ensemble mean rocessing utilizing automatic gain otimization through system arameters. This technique allows a reduction of the hysical size of the hardware and simlifies gain adjustment by reducing the gain range substantially. Such a system is not subject to Doler ultrasound system artifacts such as the mirror effect and quantization. Keywords: Doler ultrasound system, Dynamic range, Gain comensation, Mirror effect, Quantization, Ensemble mean rocess PACS number: Yb [doi: /ast.30.67] 1. INTRODUCTION The dynamic range of conventional Doler ultrasound diagnostic systems is insufficient for some alications. The blood flow signal is not sufficiently sensitive, and artifacts caused by signal clutter degrade the Doler image quality [1,2]. In recent years, diagnostic ultrasound equiment erformance has imroved through the use of high-frequency electronics and integrated circuits. New diagnostic techniques are used because of the higher dynamic range available in both B-mode imaging and blood flow analysis [3 7]. However, a higher dynamic range also requires more comlicated gain control. It is ossible to automatically otimize gain through the use of ultrasound system arameters. This technique reduces the size of the hardware and reduces the gain control range substantially. 2. DOPPLER SIGNAL PROCESSING AND ITS DYNAMIC RANGE 2.1. Signal Processing Conventional Doler ultrasound signal rocessing is shown in Fig. 1. The transceiver rocessor (Tx/Rx Proc.) receives signals from the robe elements and the received signals are amlified by the reamlifier. Gain comensation is alied to the signal to correct it for range-distance attenuation (STC: sensitivity time control), and an analog gain correction of the robe characteristics (frequency, sensitivity, etc.) is alied. The signal is then sent to an analog/digital converter (ADC). After AD conversion a baba@us.nasu.toshiba.co.j digital beam former (DBF) alies a delay attern to the data to focus it and roduce beam data. This data is rocessed by tomogram rocessing (B-Mode Image Proc.) and Doler signal rocessing (Doler Image Proc.), then is dislayed as a tomogram image and/or a sectrum Doler image in the dislay rocessor (Dislay). The signal-to-noise ratio (S/N ratio) is increased in the DBF as the number of channels (corresonding to transducer elements) to which delay calculations are alied increases. In Doler signal rocessing, quadrature detection (Mixer) is alied to the DBF outut, and a band-ass filter (BPF) rovides band limitation and rejects clutter. The result is a base-band Doler signal. At this time, the S/N ratio is sharly increased because of the band limitation of the BPF. In ulse-wave Doler signal rocessing, we aly range-gate (RG) integration across the range direction of the region of interest (ROI). This also increases the S/N ratio. In the case of a continuous-wave Doler signal rocessing, the dynamic range is even larger and the high-ass filter (HPF) alied to this data must be more sohisticated. The dynamic range of the signal leaving the HPF is also much larger than that in the case of ulse-wave Doler, on the order of 100 db [8]. After fast Fourier transform (FFT) the S/N ratio is greatly increased because of the butterfly integration. The dynamic range of the signal is now very large, and considerable gain adjustment and dislay comression must be erformed in order to dislay the data Dynamic and S/N Ratio It is also necessary to take into consideration the fact 67

2 Tx Proc. Probe that the dynamic range is increased in Doler signal rocessing. In addition, a mirror effect and/or quantization artifacts are introduced when erforming automatic gain comensation. Although the beam data leaving the DBF has a frequency of about MHz, it is resamled at about khz. Thus, the inut dynamic range of the FFT is increased by the band limitation effect. Moreover the S/N ratio of the FFT outut is increased in a similar manner to that resulting from ensemble mean rocessing [9,10]. The standard deviation of the noise increases inversely with the square root of the samle size N according to z 2 ðtþ ¼ 1 N x 2 ðtþ: z : standard deviation of outut noise x : standard deviation of inut noise Figure 2 shows a model of the S/N ratio increase induced by ensemble mean rocessing. The integration of multile channels (N channels) by the DBF and the band limitation by Mixer/BPF of Fig. 1 behave in a similar manner and can be treated similarly using Eq. (1). If the Bit Length Tx/Rx Proc. PreAm/ Analog Gain STC Fig. 2 ADC f1 Fig. 1 FFT f3 DBF f1 N1 N3 B-Mode Image Proc. Mixer /BPF PSD/Pre- Com./MA N4 Doler Image Proc. N2 Gate f2 Digital Gain Doler ultrasound system. Max Signal Level Noise Level Inut Dynamic Dislay HPF f3 Dislay Comression Effect of the ensemble mean rocess. ð1þ Outut Dynamic ensemble mean is set to N (or the band limitation ratio is 1=N), the maximum signal level will increase in roortion ffiffiffiffi to N, but the noise level increases in roortion ffiffiffiffi to N. Thus, the S/N ratio is increased a factor of N. 3. THE PURPOSE OF GAIN ADJUSTMENT Gain adjustment corrects the diagnostic target and Doler sensitivity, and the effects resulting from differences in users skills. In addition, it comensates for variations of other equiment arameters, such as the number of summing channels in the DBF, the aodization function, the bandwidth of the BPF, the integration length of RG, the FFT number, the window function, and the number of shift additions of the ower sectrum deending on swee seed [11 13]. The maximum signal level and the noise level vary with changes in these equiment arameters. In order to realize highly sensitive Doler blood-flow diagnosis without saturation, a system with a high dynamic range must erform gain comensation taking all these arameters into account. Table 1 shows a rough estimation of the dynamic range and S/N ratio based on the virtual system. A model of the signal rocessing method accomanied by an increase in the S/N ratio is shown in Fig. 3. The noise level and maximum signal level of the incoming signal are increased by signal rocessing. However, the increase in the noise level differs from that in the maximum signal level, and the overall S/N ratio is increased. Under otimal gain adjustment (the range shown in light gray in Fig. 3), there is no saturation of the maximum signal level, and the quantizing noise and the signal are not mixed in the outut. When gain adjustment is unsuitable (the range shown in dark gray in Fig. 3), a mirror effect or quantization artifacts occur in the sectrum image due to saturation or omission. For signal amlitude, when the quantization accuracy is inadequate, quantizing noise is mixed with the signal [14 16]. A method of gain adjustment that takes the margin into consideration when detecting the weak Doler signal around system noise and has sufficient quantizing margin is required. The effect of quantization is shown in Fig. 3, which shows the sectrum obtained when inutting a sinusoidal signal (0:02f s ) including white noise. The horizontal axis is time and the vertical axis is frequency normalized by the samling frequency f s. The quantizing level of the inut range was changed every 2 s and was assigned values of 3, 5, 9 and 17. It turns out that the harmonic comonents ( 20 to 30 db) were roduced by quantization with frequencies near 0:3f s, 0:2f s, and þ0:25f s. The mirror effect is an imaginal image symmetrically generated with a real image on both sides of a baseline [17]. In an analog system, it is mainly caused by the hase 68

3 T. BABA: GAIN OPTIMIZATION TECHNIQUE IN DOPPLER ULTRASOUND SYSTEM Table 1 Estimation of the increases in S/N ratio and digital gain control range. Module Cause Effect of D.R. increment 1 Conventional system Fig. 4 (db) New system Fig. 4 (db) ADC outut (Analog gain) 50 db 1 1 DBF Beam sum effect (N 1 channel) +50 db DR2 = DR logðn 1 Þ DR2 ot = DR logð Þ N 1 Mixer/BPF Band limitation effect (f 1 =f 2 ) +30 db DR3 = DR1 + DR logð f 1 =f 2 Þ DR3 ot ffiffiffiffiffiffiffiffiffi = DR1 + DR2 ot + 20 logð ð f 1 =f 2 ÞÞ RG RG integration effect (N 2 ta) +40 db DR4 = DR1 + DR2 + DR logðn 2 Þ DR4 ot = DR1 + DR2 ot + DR3 ot + 20 logð Þ N 2 FFT FFT rocessing effect [FFT number and window] (N 3 samling) +50 db DR5 = DR1 + DR2 + DR3 + DR logðn 3 Þ DR5 ot = DR1 + DR2 ot + DR3 ot + DR4 ot + 20 logð Þ N 3 PSD/Pre-Comres. Power dimension to amlitude dimension DR5 DR5 ot MA Moving average effect (N 4 average) +10 db DR6 = DR1 + DR2 + DR3 + DR4 + DR logðn 4 Þ DR6 ot = DR1 + DR2 ot + DR3 ot + DR4 ot + DR5 ot + 20 logð Þ N 4 Digital gain inut D.R. before digital gain inut 230 db (DR6) 140 db (DR6 ot) Digital gain Digital gain outut D.R. after digital gain outut (DR7) (DR7) Gain control range 1 This estimation is based on virtual model of the Doler ultrasound system. 160 db (DR6-DR7) (DR6 ot-dr7) error of quadrature detection or by small gain differences between IQ signals. In a digital system, although these effects do not occur, they are generated by a mirror effect owing to saturation. As shown in Fig. 3(c), a symmetrical mirror effect occurs on both sides of 0 Hz in the sectrum image. Figure 3(c) shows the image of sectrum in which the gain is raised 6 db every 2 s u to the sinusoidal inut including white noise. In this figure the FFT inut dynamic range is 16 bit. The horizontal axis is time and the vertical axis is the frequency normalized by the samling frequency f s. The mirror-effect comonent ( 0:2f s ) of the original signal comonent (þ0:2f s ) occurs as a result of saturation. Thus, in conventional ultrasound design, both the mirror and quantization artifacts are caused by insufficient dynamic range of the system during Doler signal rocessing. 4. DESIGN OF SIGNAL PROCESSING SYSTEM Table 1 deicts the increases in the dynamic range and gain-control range of the conventional system and the new system based on the signal rocessing block diagram shown in Fig. 1. The DBF sums the beams of the N 1 channels. The Mixer/BPF limits the band-width into f 2 =f 1. The RG integrates N 2 tas. The FFT integrates N 3 tas weighted by the window and oerator. The PSD/Pre- Com. simly transfers the dimension (from amlitude to ower) by alying the square root. The MA calculates the moving average of N 4 columns according to the swee seed of the sectrum dislay. Figure 4 shows the gain charts of the conventional system and new system based on Table 1. The gain chart of the conventional system, which does not take the realization scale of the hardware into consideration, is shown in Fig. 4. In this study, we develoed a system that can reduce the gain-control range as well as the internal dynamic range. Automatic gain comensation according to changes in the ranges of system arameters is realized for every subblock of the Doler signal rocessing and is accomanied by an imroved S/N ratio. Since the ranges of the system arameters are known, the imrovement of the S/N ratio, the maximum signal level, and the noise level are calculable. The internal dynamic range and the gain adjustment range can be otimally designed for every subblock. By connecting artially otimized subblocks in series and uniting them, the internal dynamic range of the system can be reduced, thus, the system size and the total gain control range can both be sharly reduced. As shown in the examle in Fig. ffiffiffiffi 2, the internal S/N ratio is increased by a factor of N. Then, suosing the inut signal 69

4 Bit Length Inner Dynamic Saturation Region Max. Signal Level Ot. Gain Small Gain Large Gain Noise Level Margin Otimized Gain Quantizing Noise Region Imroer Gain N=17 86dB N=9 92dB N=5 98dB N=3 (c) 104dB Fig. 3 Artifacts caused by inadequate gain control. Problems of inadequate gain control. Artifacts caused by quantization. (c) Artifacts caused by mirror effect. dynamic ffiffiffiffi range is DRin [db], a range increase equivalent to 20 logð N Þ [db] will occur. Moreover, the internal dynamic range ffiffiffiffiffi DRroc [db], adds a margin of more than (20 logð 12 Þ) to the quantizing noise, can be roughly calculated using ffiffiffiffi DRroc DRin þ 20 logð N Þþ20 logð 12 Þ: ð2þ When digitizing, the least significant bit (LSB) must be rounded u or down not truncated, otherwise an error of LSB=2 will occur. The RMS value of the quantizing noise is equivalent to LSB= ffiffiffiffiffi 12. Thus, an additional margin of 12 for the dynamic range must be maintained so that rounding can be erformed accurately [18]. Although the internal dynamic range DRroc is only sufficient for automatic gain comensation using the system arameters, it is necessary to consider a margin for the original gain adjustment that comensates for the diagnostic target and Doler sensitivity variations in an internal dynamic range. The gain chart of the subblock signal rocessing accomanied by range exansion is shown in Fig. 4. The gaincontrol range of the conventional system (Fig. 4) is DR6-DR7, and the gain-control range of the new system (Fig. 4) is DR6 ot-dr7. In general Doler signal Fig. 4 Comarison of gain control systems. Gain chart of the conventional system. Gain chart of the new system. rocessing, DR6 is above 200 db and DR7 (the digital gain outut) is the dislay luminance range (about ). Thus, the gain-control range of the conventional system should be more than 130 db, which is very large. On the other hand, DR6 ot for new system is smaller than DR6, about 100 db. We can reduce not only the gain-control range but also the inner dynamic ranges of the submodules at the same time. 5. RESULT The effect of automatic gain otimization using the new system was verified in a simulation of RG integral rocessing. Sectra of the conventional system and the new system when changing the RG width are shown in Figs. 5 and 5, resectively. The horizontal axis is time and the vertical axis is the frequency normalized by samling frequency f s. The RG was adjusted from 1 mm to 4 mm then to 16 mm in 1.8 s intervals. A sinusoidal signal including white noise was used as the inut. In the 70

5 T. BABA: GAIN OPTIMIZATION TECHNIQUE IN DOPPLER ULTRASOUND SYSTEM 1mm 4mm 16mm 1mm conventional system, the signal level and noise level increase with increasing RG width. For this reason, the user should reduce gain manually when the RG width is increased. In the new system, although the signal level rises if the RG width is increases, it turns out that the noise level does not change. As mentioned above, by Doler automatic gain comensation, the inut bit length of each signal rocessing block can be made smaller, and the gain adjustment range can also be decreased as necessary. 6. CONCLUSION Through the use of ensemble mean rocessing it is ossible to reduce the overall dynamic range of an ultrasound signal while maintaining signal quality. As a result, the bit deth required to reresent the signal can be reduced, resulting in smaller and more cost-effective system hardware. Furthermore, gain adjustments of the oerator and the final system are simlified meaning that less oerator adjustment is required. The system does not generate the mirror and quantization artifacts that occur in conventional ultrasound system designs. REFERENCES 4mm Fig. 5 Effect of automatic gain comensation: examle of RG integration rocess. Conventional system. New system. 16mm [1] T. Baba, Investigation of wall filter in Doler ultrasound system, Soc. Signal Process. Al. Technol. Jn., 9(2), (2006) (in Jaanese). [2] T. Baba, Evaluation of ost wall filter for Doler ultrasound system, in Acoustical Imaging, Vol. 29, I. Akiyama, Ed. (Sringer Science + Business Media, Berlin, 2008), [3] J. Lisauskas, J. Singh, M. Courtois and S. J. Kovacs, The relation of the eak Doler E-wave to eak mitral annulus velocity ratio to diastolic function, Ultrasound Med. Biol., 27, (2001). [4] C. Bruch, R. Gradaus, S. Gunia, G. Breithardt and T. Wichter, Doler tissue analysis of mitral annular velocities, J. Am. Soc. Echocardiograhy, 16, (2003). [5] C. W. Guron, M. Hartford, A. Persson, J. Herlitz, D. Thelle and K. Caidahl, Timing of regional left ventricular lengthening by ulsed tissue Doler, J. Am. Soc. Echocardiograhy, 17, (2004). [6] H. Kanai and Y. Koiwa, Myocardial raid velocity distribution, Ultrasound Med. Biol., 27, (2001). [7] T. Oki, Y. Oishi and T. Ishimoto, Tissue Doler imaging, J. Med. Ultrason., 30, (2003) (in Jaanese). [8] J. A. Jensen, Estimation of Blood Velocities Using Ultrasound: A Signal Processing Aroach (Cambridge University Press, London, 1996), [9] The Physical Society of Jaan, Exerimental Data Processing Using Comuter (Saiensu-sha Co., Ltd., Tokyo, 1987), (in Jaanese). [10] G. Date, A. V. Oenheim and R. W. Schafer, Digital Signal Processing (Corona Publishing Co. Ltd., Tokyo, 1981), (in Jaanese). [11] K. Ranganathan and W. F. Walker, A novel beamformer design method for medical ultrasound. Part I: Theory, IEEE Trans. Ultrason. Ferroelectr. Freq. Control, UFFC-50, (2003). [12] M. Nagano and T. Nakata, Frequency swee sectrum analyzer, Soc. Signal Process. Al. Technol. Jn., 3(4), (2000) (in Jaanese). [13] K. Kubo, Method for estimating instantaneous frequency and arameters of sine-wave by differential domain samling, Trans. Soc. Instrum. Control Eng., 42, (2006) (in Jaanese). [14] K. Miyawaki, Noise Analysis (Asakura Publishing Co. Ltd., Osaka, 1978), (in Jaanese). [15] Y. Kasahara, Information Theory and Communication Method (Kyoritsushuan Co., Ltd., Tokyo), 1976, (in Jaanese). [16] S. Minami and S. Kawada, The Introduction of Data Processing for Science Measurement (CQ Pub. Co., Ltd., Tokyo, 2002), (in Jaanese). [17] H. Susaki, I=Q samling method as re-rocessing: Frequency analysis, Soc. Signal Process. Al. Technol. Jn., 5(3),. 2 7 (2002) (in Jaanese). [18] V. Caellini, A. G. Constantinides and P. Emiliani, Digital Filters and Their Alications (Academic Press, London, 1978), Tatsuro Baba received the B.S., M. Sc. in recision engineering from Osaka University, and Ph.D. degree in Mechanical engineering from Kobe University in 1981, 1983 and 2006, resectively. From 1983 has been emloyed at Toshiba Medical Systems Cororation, Tochigi. His research interests include Doler ultrasound system and digital signal rocessing. 71