Last year, nearly 100 million medical ultrasound scans were performed in the U.S. The popularity of medical ultrasound is due in large part to its wide availability, relatively inexpensive cost, and absence of measurable long-term harm to patients. In an effort to sustain these benefits as ultrasound manufacturers move to deliver next-generation machines, the introduction of real-time compression into the design process will help lower the cost and complexity of ultrasound scanners.
Ultrasound machines utilize between 32 and 256 piezoelectric transceivers within the ultrasound probe, which generate from 30 to 60 acoustic pulses per second. When these pulses scan the patient’s skin, muscle, tissue, and bone, the same piezoelectric transducers record acoustic reflections and refractions at sample rates between 10 Msamples/s and 50 Msamples/s, typically using 12 bits per sample to maintain about 70-dB dynamic range.
However, since patients’ ultrasound reflections vary over an approximately 100-dB dynamic range, additional variable gain amplifiers (VGAs) are inserted prior to analog-to-digital conversion to provide up to 30 dB of analog gain during the later stages of pulse recording. Digitized transducer channels are then fed to a beamformer (see the figure), which coherently combines multiple transducer channels arriving from a particular direction (steering angle) to create a focused beam per ultrasound transmit pulse.
Because each beam only represents a small angle in the field of view, between 32 and 128 beams must be combined to create each frame of data. Typical diagnostic ultrasound machines operate at 30 frames/s, although frame rates for cardiac applications may be as high as 60 frames/s.
A “back of the envelope” calculation demonstrates that analog-to-digital converters (ADCs) in ultrasound front ends generate prodigious amounts of sampled data: 256 channels x 50 Msamples/s x 12 bits/sample = 154 Gbits/s that converge on the beamformer. As an example, 384 low-voltage differential signaling (LVDS) pins operating at 800 MHz are required just to deliver the analog-to-digital samples to the beamformer, which is typically implemented in one or more FPGAs.
The Role of Real Time
To reduce the LVDS pin count, bit rate, or both, ultrasound manufacturers have successfully been applying real-time compression to lower the cost and complexity of their ultrasound scanners. In fact, transducers using Samplify’s Prism 3 compression can achieve 2:1 lossless compression—and according to one ultrasound manufacturer, even 3:1 compression—with no measurable effect on diagnostic image quality. At 3:1 compression, the ADC connections are reduced from 384 to 128 pins, allowing medical imaging companies to use a less expensive FPGA and a smaller, more cost-effective package.
Looking forward, next-generation 4D ultrasound machines will soon display 3D images that change from frame to frame. But these 4D machines will require thousands of transducers, making the analog-to-digital-to-beamformer interconnect problem an order of magnitude more bandwidth-intensive than today’s machines. For 4D machines, compression of the beamformer output signals will be required to use serializer-deserializer (SERDES) interconnects like PCI Express Gen2 and Gen3 and USB 3.0.
Ultrasound engineers familiar with FPGA pricing structures are quickly realizing the economic benefits of using real-time compression and decompression with lower-cost Altera Cyclone IV or Xilinx Spartan 6 devices with 3-Gbit/s transceivers, rather than using the significantly more expensive 10-Gbit/s SERDES transceivers found on Altera Stratix IV or Xilinx Virtex 6 devices.