4. Theory Ref.
β’ Synthetic Aperture Ultrasound Imaging
4
Jensen J., Nikolov S., Gammelmark K., Pedersen M.
Ultrasonics 44 (2006) e5βe15
Technical University of Denmark, Center for Fast Ultrasound Imaging
5. Conventional ultrasound imaging
5
β’ Waves travel through tissue and partly
reflected at each tissue interface
β’ B-mode: multiple lines of interrogation over
a wide area, each returning echo is assigned
β a brightness on a grey scale based on Intensity β
6. Beam forming
6
β’ Form the shape, size and position of the ultrasound beams
β’ Controlling by generation of electrical signals
9. Dynamic Reception Focusing
β’ Reception focusing
β’ Change electronic focusing delays with time
β’ Dynamic delays
β’ At time T focus is at depth(ππ/π)
9
9
10. β’ Axial resolution frequency.
β’ Lateral resolution Lambda - aperture size - depth
β’ The acquisition rate speed of sound c
Aperture synthesis
β
β
β
10
11. β’ One image line at a time
β’ Resolution :
Aperture synthesis
11
βπ₯ =
Ξ»
πΏ
π§π
12. Conventional Imaging Limits
12
Impossible to focus every where in the image
Dependence on Wave Length β Focul depth β Aperture size
One Image line at a time.
Focus in Transmission
15. Synthetic Aperture
15
1950s
Originally conceived for Radar systems
1970s
Initially implemented using digital computers.
Radar β Ultrasound Applications
1990s
Medical US imaging
Intravscular imaging.
2006
Jensen Paper
Implementaion
Real Time SAU imaging
2020
-------------------------
GPU β FPGA
18. Synthetic Aperture vs Conventional
β’ Single element is used for transmitting a spherical wave
covering the full image region.
β’ The received signals can be used for making a low
resolution image.
β’ Focusing is performed by finding the geometric
distance from the transmitting element to the imaging
point and back to the receiving element
β’ When a short pulse is transmitted and the echo signal
is received, a round-trip delay is:
18
19. β’ Resolution :
Synthetic Aperture vs Conventional
17
SAU
Resolution is constant everywhere with the same aperture size.
Conventional
The best resolution is only available at the focal depth
π³ π = πππππ
βπ½
π
β π. βπ½
βπ₯π π¦ππ‘β.ππ β
πΏ
2 βπ₯ =
Ξ»
πΏ
π§π
21. High Resolution Image
π»π πΌ ππ =
π=1
ππ₯ππ
π=1
ππΈ
π π‘π π, π , π, π π¦π(π‘π π, π , π, π)
21
Number of transducer
elements
Number of emissions
Geometric distance
from emitting element to
the imaging point
p =1,2,3,β¦ P
the current point in the
set of P points
Apodization function for
emission i on transducer j
Received signal from
emission i on transducer j
time for transmission and reception. Function of transmitting (i)
element position, direction of propagation (i) and receiving (j)
element position
25. Final Flow Chart
25
β’ The recorded wave field is first Fourier
transformed.
β’ For every depth π§ of interest, the wave field is
extrapolated to π§
β’ Integrated over π€, and inverse Fourier
transformed, producing a focused image line
ππ π₯, π§
25
26. Penetration problem
β’ Single Element
β’ Solution : Combining several elements for transmission and
using longer waveforms emitting more energy.
26
27. Flow Estimation
β’ In SA imaging, it is possible to focus the
received data in any direction and in any
order. It does not have to be along the
direction of the emitted beam, since the
emission is spherical and illuminates the
full region of interest.
27
29. Angular Resolution
β
40 β
30 β
20 β10 0 10 20 30 40
0
β10
β20
β30
β40
β50
β60
β70
Level
[dB]
boxcar/hanning
β30
β40
β50
β60
β70
β20
β10
0
Level
[dB]
β40 β30 β20 β10 10 20 30 40
boxcar/boxcar
2 emissions
4 emissions
8 emissions
16 emissions
32 emissions
64 emissions
Angular resolution of a SA imaging system for diο¬erent number of emissions, when using a boxcar apodization in transmit and a boxcar (bottom) or
Hanning apodization in receive (top)
Jensen J., Nikolov S., Gammelmark K., Pedersen M. Ultrasonics 44 (2006)e5βe15
29
30. Low resolution images combined to produce a high resolution image. One element transmit at the time, while all are used to receive.
The images are then added into one high resolution image
Field II simulation Results
3030
32. Attempt for implementation
β’ First Official Device : RASMUS
β’ All conventional US imaging methods can
be implemented with this system, but
Real-time SA imaging is not possible.
32
34. Implementation
Amount of Calculations of SAU
1) Transmit β Receive :
2) Focusing delay
3) Apodization value
4) Interpolating the sample value
5)Summing to other values
Nc=kNl Ne 4f0
3434
ππ = 200 ππ = 192 π0 = 5ππ»π§ π = 0.8
36. Xilinx Versal ACAP
36
β’SA and PW beamformer on Versalβ’ for
Ultrafast B-mode and Flow imaging
β’Associated C++ and Python classes to
produce the SA and PW design
β’Frame rates in the range of 1000 fps in a 64
channel architecture on a single Versal using
50% of the available AI Engines in Versal
36
37. AI Engine : Tensor Representation
Vector
(Tensor 1D)
1
Scanline
Matrix
(Tensor 2D)
128
Scanlines
Cube
(Tensor 3D)
128
Scanlines
128
Cube of Cubes
(Tensor 4D)
128
Scanlines
1200
Sample
s
1200
Sample
s
1200
Sample
s
1200
Sample
s
3737
39. 4 Kernels
Beamforming Image
Front End
Programmable
Logic
AI Engine to
DDR Mover
Calculate the
Points for a
Scan Line
Calculate
Focused Field for
Virtual Sources
Control and Scheduling
Transducer
RF Data
Find Transit
Delay for the
Line
Dynamic
Apodization
Add Receive
Delay Time to
Transmit Delay
Calculate
Apodization
Interpolate
Samples
Sum Interpolated
Samples by
Apodization
For Each Line
128 Lines
Processor
Cortex-A72
For All Elements
32 Elements
Calculate Calculate
Apodization FiFo Delays
iFo
F
Graph Structure
3939