Introduction to Ultrafast Ultrasound Imaging

Ultrasound Imaging is a specific type of Medical Imaging where we want to drive ultrasound waves in order to obtain information from the human body. This information is usually displayed as an image with different brightness levels depending on how we receive information from the back-scatter of the emitted wave in the human body (the so-called B-Mode).

Classical ultrasound imaging relies on focusing a specific point in the human body, at a very specific depth, and then reconstructing the B-Mode image line by line (in a vertical manner) based on the received echoes from the diffraction process of the emitted waves. Following a little diagram of it:

As it can be seen, the image is obtained by having all the elements of the ultrasound probe focusing the emission on a single “focal area”. This methodology, although it provides a robust imaging modality, it lacks of different qualities required from modern systems:

●?????? It has a very limited frame rate, as the image formation is limited by the speed of sound in the body (we can approximate it to ~1540 m/s), the number of lines in the image and the inspection depth

○?????? The more lines and more depthness causes the limit of frames to significantly drop

●?????? Although it can be robust for many diagnostic purposes, it is not desirable for cardiac imaging or therapeutical ultrasound (i.e. microbubbles treatment)

For this reason, during past and recent years, many steps have been taken to overcome those limits, both from an imaging perspective and technology. Let’s focus for the moment on imaging modality.

During the course of years, many imaging modalities such as “multi-line” beamforming or synthetic focus beamforming have been introduced. The latter, synthetic focus beamforming is able to ideally overcome most of the problems detailed in the previous paragraph. In fact synthetic focus beamforming emits ultrasound waves across the whole region available from the probe, and reconstructs digitally the focus of the emitted wave. The two main imaging modalities are called Synthetic Aperture [REF] and Plane Wave. Following a little scheme of the two emission modalities:

Synthetic Aperture (SA) is inspired by military RADAR, where we synthetically open the emission by exciting sub-groups of elements to emit a spherical wave and then receive with all the elements. The focus is reconstructed digitally and can be done both in transmission and reception. The reconstructed images from the emission of the sub-groups of elements are then compounded together in order to produce the final image.

Plane Wave (PW), on the contrary, emits a plane wave by exciting all the transducers at the same time, and steers the wave emitted with different angles. As per SA, PW compounds coherently the emitted waves to form a single image.

Advantages of Synthetic Focus Beamforming

Synthetic Focus Beamforming overcome most of the classical problems, bringing many advantages:

●?????? Removes the limitation of frame rate

●?????? Coherent summation and/or interpolation of multiple emissions produce a final image with extremely low Signal-to-Noise ratio

●?????? Ideal for any inspection depth

●?????? Very good axial - lateral dispersion

But how can Synthetic Focus Beamforming work so well if we need to reconstruct everything digitally? Let’s then focus on spatio-temporal coherency and inspection depth.

The first version of synthetic focus beamforming, especially SA, had a very consistent problem, or rather the inspection depth.

SA in fact, in first formulations, had a penetration problem, which can be partially recovered with the number of emissions, but still can cause spatial inconsistencies if no manipulation happens. To eliminate both problems of spatial consistency and depth of inspection, virtual sources were introduced.

A virtual source, or virtual source of emission, is an imaginary point in space which receives the echoes from a previous point and emits them in the direction of the set depth inspection. The formulation of virtual sources allows one to create an imaginary spatial grid, which covers the space of emission and create a virtual “image” in which the net of sources, can be “arbitrarily” grown in function of some sampling and emission parameters. It is implied that this structure provides an extremely consistent spatial resolution. Following a diagram of how virtual sources works:

Is Synthetic Focus Beamforming computable?

One question that always arises when discussing Synthetic Focus Beamforming is: Is the algorithm computable with the appropriate time constraints?

This question has its root in the complexity of the algorithm itself, which requires to digitally reconstruct the focus, compute on the fly the dynamic apodization window, interpolate the samples and accumulate all the results over the scanline per single emission done. This process must be repeated for all the emissions and the emissions summed coherently and interpolated. The result of beamform per emission is called Low Resolution Image (because it has a poor SNR index), whereas the final image, result of the coherent sums is called High Resolution Image, as the coherent summation removes most of the noise contained in the LRIs. Following a schema of the algorithm:

With traditional silicon hardware, it is extremely difficult to create a synthetic focus system, as the number of Multiply-Accumulate (MAC or MACC) operations and reordering is enormous.

However, with AMD Versal, it is possible to achieve a sufficient number of operations, both from MACC, memory operations and reordering operation standpoint. Moreover, using? AMD Versal, it is possible to fit everything in a single chip, reducing exponentially the hardware complexity.

Just to provide a rough estimation, the algorithm must be able to parse data in the order of 10-15 GB/s, with a required computational power of 1-30 TOPs/s, depending on the inspection depth and the number of emission of the Synthetic Beamforming Focusing. Aside from the hundred of TOPs/s available in real-time, AMD Versal has two tremendous advantages:

1. It has up to 70MB of memory in the fabric and accelerators, allowing to parse enormous amount of data per second
2. It has a Network-on-Chip (NoC), which allows to transfer data with magnitudes of several Terabit/s

We will cover an implementation in the next articles, however please note that both SA and PW are both data-intensive and compute-intensive algorithms, and AMD Versal solves all of these problems at once.

Conclusion

We have introduced Synthetic Focus Beamforming, and its two imaging modalities, namely Synthetic Aperture and Plane Wave. We have introduced the advantages with respect to other and conventional imaging techniques. MakarenaLabs, due to its capabilities and leveraging partner’s technology such as AMD Versal, is able to create a full ultrasound system with the most advanced imaging and flow imaging modalities on the same chip. The last advantage of implementing these algorithms on hardwares such as? AMD Versal, is the safety and security of the system. In fact, all the silicon listed is able to provide support and certifiability up to SIL3, making it suitable for authorities clearance and approval of the final system even for the most critical applications (i.e. during diagnostic of first aid operation).

In the next articles we will cover how to implement a premium performance beamformer on a single chip and how to add AI capabilities to it.