Ultrasound generally is considered a very safe type of nonintrusive medical scanning. It doesn’t produce the ionizing radiation associated with X-ray imaging, nor does it require the expense and size of an MRI installation. Portable ultrasound machines are manufactured by many companies, and a number of distinct operating modes have been developed. Because it is safe and relatively low cost, ultrasound scans have become a routine part of many medical procedures.
The B scan mode—B for brightness—is the basic imaging capability. In addition, Doppler functionality in some machines overlays a color-coded indication of blood-flow speed through arteries and veins within the region of interest.
Except for the continuous Doppler mode, pulses of high-frequency sound waves are applied to the relevant area of the body, and the reflected echoes are used to generate an image. High frequency for ultrasound is from 1 MHz to 15 MHz or higher. Doppler can be pulsed, but it also may be implemented as a continuous wave.
Although basic analog ultrasound imaging began about 50 years ago, today’s machines are largely digital and provide many more options as well as images with much greater detail and color. Most modern equipment is based on phased array scanning similar to advanced sonar and radar, and it all requires very high performance components.
As an example of the circuitry involved, National Semiconductor (now part of Texas Instruments) has developed an eight-channel Ultrasound Tx/Rx Chipset, the LM965XX series of parts. It comprises a digital transmit beamformer, a high-voltage transmit/receive (Tx/Rx) switch, a high-voltage ultrasound Tx pulser, and an analog front-end (AFE) IC.
For ultrasound applications, a channel is associated with each of the several elements that together form the transducer a doctor holds against a patient’s body. This chipset provides eight high-power pulse inputs to eight elements and develops eight digitized AFE received echo outputs.
The recently introduced United Imaging Healthcare iuStar100 Ultrasound Machine has 48 channels, which is near the minimum quantity in commercial machines. Transducers may comprise 100 or more elements, and typically, each has its own dedicated Tx/Rx electronics. Although these systems require a lot of specialized circuitry because of their nature, integration by National, Maxim Integrated Products, Samplify Systems, Analog Devices, and others has greatly affected the cost, portability, and performance of modern medical ultrasound machines. The iuStar100 is based on Samplify electronics.
The National Ultrasound Chipset primarily is analog in nature other than the Tx beamformer timing. Indeed, the success of a phased array ultrasound design depends on the accuracy with which the elements can be steered to produce a focused energy beam as well as the noise level that can be achieved for the low-level return signals. Assuming both aspects are optimized, the major remaining job is to condition and combine the digitized return signals to form an image.
As succinctly noted in a Texas Instruments (TI) white paper,1 “Ultrasound systems are signal-processing intensive.” To put that into perspective, a recent academic paper proposed using two TI eight-core 1.25-GHz C6678 DSP ICs to compute basic beam steering and beamforming algorithms for a relatively small 64-channel ultrasound system. Nevertheless, this was an important contribution because most solutions use FPGAs and/or ASICs to deal with the amount, speed, and complexity of the signal processing.2
Transmit beam steering refers to the coordination of the individual element pulses so the energy from them simultaneously arrives at the desired point. Physically, the elements are fixed in their relation to one another. They may be glued together to form a simple linear probe or angularly offset one from the next in a shallow fan shape. Regardless of the exact arrangement, beam steering is done electronically by suitably timing the individual pulse bursts with respect to each other. The goal is to approximate a concave wavefront focused on a point at the desired depth and position along a radial scan line (Figure 1).
Figure 1. Focusing Ultrasound From Transducer Elements
Courtesy of Texas Instruments
A uniform tissue impedance is assumed, corresponding to 1.54-km/s ultrasound propagation speed or 1.54-mm/µs. So, sub-mm resolution requires sub-µs timing. The National LM96570 Tx beamformer IC is very flexible and supports individually programmable 64 pulse patterns with delays up to 102.4 µs adjustable in 0.78-ns increments. Pulses are output at up to an 80-MHz rate.
A basic system would simply output a burst of pulses on each of several contiguous elements, with each burst offset slightly from the next. The required delays are derived from geometric considerations. For example, to focus at a point toward the right-hand end of a linear transducer array, the pulses originating from the left end must travel farther than those nearer on the right.
Unfortunately, human tissue does not present the same minimal scattering environment as does the water sonar passes through or even the air that radar encounters. In a paper on receive signal processing, the authors comment on the properties of human tissue: “The medium will reflect at all angles and at all depths at the same time, and there are hardly any nonreflecting regions at all as in e.g. underwater sonar.”3
In addition, the propagation speed in actual tissue differs slightly from the nominal, which affects the beam focus. To further complicate things, tissue attenuates ultrasound at approximately 1.4-dB/cm/MHz round trip. In other words, a 10-MHz signal imaging a point 1 cm away from the transducer element has a 2-cm round trip and would be attenuated 28 dB when received by the same element.
The simplest approach is to drive an element with a square wave at a certain frequency. The fundamental contains more energy than any of the individual harmonics, so the idea is to minimize contributions from the harmonics in both the transmit and receive modes of operation. High-performance ultrasound systems shape the transmitted pulses to minimize the second harmonic content.
This can be done using an Arb but at large expense. Typically, newer systems have transferred the function to sophisticated pulser implementations that quickly switch among several power supply levels and ground to directly generate a multilevel pulse. For example, the Maxim MAX4940/4940A is a quad, 2.1-A unipolar/bipolar pulser that can switch from +110 V to -110 V and from either supply to ground. If only a unipolar pulse is required, the maximum voltage doubles to +220 V.
For this device operating at a 5-MHz output frequency, the second harmonic is at least 40 dB below the fundamental. This is important for a couple of reasons. An ideal square wave has only odd harmonics, so the level of the second harmonic is an indication of waveform distortion. Also, because image resolution is related to wavelength, some ultrasound machines are capable of detecting the second harmonic to improve resolution.
Human body tissue is nonlinear and actually generates a second harmonic signal in response to energy at the fundamental. Obviously, this is the signal you want to detect, not the distortion present in the transmitted pulse.
Focusing is largely based on geometry in a basic phased array system. You don’t know beforehand what the effect of different tissue layers might be. For example, a 0.5% change in the speed of sound is a difference of 0.0077 mm/µs. Focusing on a point at a 1.54-cm depth would have a 20-µs round-trip time if the speed of sound were exactly 1.54 mm/µs. If it is off by 0.5%, the depth of the focused point will be off by about ±0.15 mm.
Looking at focusing another way, how well an ideal wavefront can be approximated is governed by how closely the transducer elements can be spaced and the timing resolution of their various delays. A 10-ns resolution corresponds to 0.0154 mm, so it would appear that accurate timing may be readily achieved.
Indeed, the National LM96570 delays have 0.75-ns resolution and Samplify’s SFF9140 Transmit Beamformer FPGA 5-ns when clocked at 200 MHz. Actually, the National part operates with a basic 6.25-ns period associated with an on-chip PLL-derived eight-phase 160-MHz clock. Coarse delays count clocks, and fine delays select the appropriate phase.
Timing resolution in the receive signal chain is limited by the ADC sampling rate—typically 50 MHz. An ADC running at 50 MS/s has a 20-ns sample period, so the focus uncertainty is about 0.03 mm.
Several steps are taken to reduce errors. Apodization is applied during signal reception and transmission to improve the SNR by suitably shaping the Tx or Rx beam. In addition, the number of elements being used is changed depending on the focus depth to maintain similar lateral resolution at all depths.
For greater focus depths, all transducer elements may be used with Tx apodization accomplished by proportionately reducing the drive level to the elements farther from the center of the transducer. This approach reduces side lobes because it changes the aperture shape from rectangular (unweighted) to approximate something smoother such as a Gaussian, Hamming, or Hanning function.
For a rectangular aperture, the far-field beam pattern is a sinc function. So, if the depth of field is too large and a rectangular Tx aperture is used, the sinc-function side lobes could allow a reflection from beyond the intended focus depth to be confused with the desired signal.4
Reduction in the number of elements narrows the effective beam aperture, which increases the F number for a given focus depth. The F number equals focus depth/aperture and is variable. The width of the focused beam increases with the F number as does the depth of field. By appropriately setting the delay for each element, a good focus can be developed at any reasonable depth. But, how good is good?
For ultrasound imaging, a small F number is desirable because it corresponds to good lateral resolution. According to a GE ultrasound technology update paper, the company’s LOGIQ® 700 MR Machine combines both low Tx and Rx F numbers to significantly improve lateral resolution. Conventional machines often transmit with F number between 2 and 3 to cover a larger area. The receive F number can be reduced by operating with a larger receive aperture.5
As the various adjustments are being made to the Tx configuration for each point along each scan line, a corresponding set of variables must change in-step on the receive side. The basic requirement is to delay received signals depending on the Tx delays, so the received signals become time-aligned. Beyond that, the correct elements much be chosen to achieve the desired image characteristics.
In the ultrasound technology update, GE claims that, with a 128-channel beamformer, the 700 MR Machine produces images using Tx and Rx F numbers less than 1.0. The improvement in lateral resolution can be as large as 4x.
The Receive Signal Path
Signal absorption is about 0.5 dB/cm/MHz in human tissue. Although operating at the second harmonic provides better resolution, because of higher absorption, the usable depth is limited. Depth-variable absorption also means that the receive gain should vary with focus depth. This is accomplished by the time-gain-amplifier (TGA). Time is involved only because points along a scan line are focused at successive time increments.
The TGA actually is adjusted to track the attenuation associated with different depths. As an example, the Analog Devices AD9276/77 Octal Ultrasound Receivers have a 42-dB variable gain range. The Maxim MAX2077 Octal-Channel Ultrasound Front End has a separate low-noise amplifier (LNA) ahead of a programmable gain amplifier. The LNA has both 12.5-dB low gain and 18.5-dB high gain modes to extend the system dynamic range.
It has only been fairly recently that Doppler signals could be handled by the same receive electronics as normal reflections. One reason amplifier design is difficult is the need for very low near-carrier modulation noise. Low-velocity pulsed and color-flow Doppler sensitivity must be high to detect low levels of modulation. The AD9276/77 offers 16 values of programmable phase rotation for CW Doppler operation.
The AD9276/77 and MAX2077 include an anti-alias filter on each output. National’s LM965XX chipset is partitioned differently so the LM96511 Ultrasound AFE includes 12-bit ADCs but separately outputs the raw Doppler signal from the LNA.
Samplify has taken yet another approach. The company has developed the 32-channel SMM9132/33 Ultrasound AFE Receiver Modules, which are complete ultrasound receivers in a SO-DIMM format. To get started, a company wishing to develop an ultrasound machine only has to add post-processing after this module.
According to Danny Kreindler, Samplify’s director of ultrasound product marketing, hundreds of application examples are shipped with the modules so customers have a good starting point for almost anything they might want to do. Few restrictions have been embedded in the system so the modules also are popular with research institutions because of their flexibility. Two of Samplify’s low-power SAM1600 12-bit, 16-channel ADCs are included in the modules along with four Maxim MAX2077 AFEs and a complete set of 32 Tx/Rx switches.
The SMK9130 is Samplify’s 64-channel development kit and includes two of the SMM9132/33 modules, two 32-channel SAM2032 AutoFocusTM Beamformer ASICs, the SFF9130 Mid-Processor/Host Interface FPGA, the 64-channel SFF9140 Transmit Beamformer FPGA, and a couple of SMM9152 HV Power Supply Modules (Figure 2).
Figure 2. SMK9130 AutoFocus Beamforming Front-End Block Diagram
Courtesy of Samplify Systems
Standard post-processing includes the R-q conversion to rectangular coordinates from the polar image comprising the original scan lines. However, there are several areas in which further processing can improve performance.
For example, once the received signal has been digitized, the time alignment is only as good as the ADC sample period. Instead, some systems upconvert the ADC output via interpolating filters to improve resolution by as much as 4x. In addition, just as the transmit beam was shaped or apodized, so too the received channel signals can be weighted before summing. These steps reduce the side-lobe interference and improve the final image quality.
Further processing is necessary to merge Doppler information with the basic image data. Although most of the operating modes have become standardized, each manufacturer has special features.
Innovative Ultrasound Systems
Among the many medical ultrasound system manufacturers, GE and Philips Electronics are two of the largest. They are listed in a recent article about Samsung Electronics’ purchase of Korea-based Medison with approximately 7% of the world’s ultrasound diagnostics market. Medison’s technologies support innovative 3-D and 4-D systems in addition to more conventional ones.6
According to a Medison datasheet, high-definition volume imaging (HDVITM) “is based on nonstationary adaptive filtering, resulting in removal of unwanted speckle and noise whilst increasing visualization of edges and small structures in volume data. HDVI uses a 3-D processing algorithm based on 3-D matrix processing of volume data.”7 The technology is used in the company’s ACCUVIX systems.
GE’s B-Flow technology in the LOGIC systems enhances the receive-side sensitivity through correlation techniques. As described in a GE paper, “The digitally encoded ultrasound (DEU) beamformer consists of a transmit encoder and a receive decoder in addition to array focusing electronics. B-Flow uses…coded excitation to send coded pulse sequences into the body. It then decodes the returning signals to enhance sensitivity to weak signals such as blood cells.” B-Flow isn’t based on Doppler but instead subtracts successive images, displaying the moving parts such as blood flow and suppressing the stationary parts.8
Philips has developed the xMATRIX array technology that features an integrated X-Y transducer with thousands of elements. With the 2,400-element iE33 transducer, beams can be steered and focused through the entire 360 degrees to provide the best view of difficult-to-image areas such as parts of the heart.
The Live 3D TEE (transesophageal echo) mode gives cardiologists views of cardiac structure and function seen for the first time. You can view the 3-D heart in real time as well as review the 3-D volume at any time, select any plane for interrogation, and perform extensive quantification.
The company’s iU22 transducer with 9,212 elements trades resolution for speed and is more suitable for abdominal, obstetrical, fetal echo, and gynecological applications (Figure 3).9
Finally, Siemens is promoting the ACUSON S2000 Machine that measures the relative stiffness of tissues depending on how they respond to an ultrasound “push pulse.” They call it acoustic radiation force impulse imaging. The company’s Virtual TouchTM Software applications provide a numerical value for the speed of shear waves traveling through tissue in the region of interest. The shear waves travel perpendicular to the direction of the push pulse, and shear wave speed correlates to tissue stiffness.10
Figure 3. 3-D View of Back Fascia Tumor Imaged With iU22 Ultrasound System
Courtesy of Philips Healthcare
- “Signal Processing Overview of Ultrasound Systems for Medical Imaging,” Texas Instruments, White Paper, 2008.
- Karadayi, K. et al, “Software-Based Ultrasound Beamforming on Multicore DSPs,” University of Washington, 2011.
- Holm, S. et al, “Capon Beamforming for Active Ultrasound Imaging Systems,” Proceedings of IEEE 13th DSP Workshop, 2009.
- Szabo, T. L., Diagnostic Ultrasound Imaging Inside Out, Elsevier Academic Press, 2004, p. 148.
- “Wide Aperture/Low F-Number Imaging: Ultrasound Technology Update,” GE Healthcare, 2007.
- “Pushing the Cutting Edge of Ultrasound Technology,” Medical Device and Diagnostic Industry, June 2011, p. 74.
- “HD Volume Imaging,” Samsung Medison, http://www.samsungmedison.com/technology/hd-volume-imaging.jsp
- Clevert, M.D., D-A., “Vascular Imaging With B-Flow,” GE Healthcare Online CME Course, 2010.
- “xMATRIX Array Transducer With PureWave Crystal Technology,” Philips Healthcare, www.healthcare.philips.com
- “Tissue Strain Analytics: Virtual Touch Tissue Imaging and Quantification, Siemens, White Paper, 2008.
For More Information