Combining spectral and intensity data to

1
Combining spectral and intensity data to
identify regions of cavitation in
ultrasound images application to HIFU
Chang-yu Hsieh1, Penny Probert Smith1, Tom
Leslie2, James Kennedy2 , Fares Mayia3 and
Gouliang Ye1


1Wolfson Medical
Vision Laboratory, Department of Engineering
Science, University of Oxford

2The HIFU Unit,
Churchill Hospital, Oxford Radcliffe Health
Trust, Oxford



3Department of Medical Physics Churchill
Hospital, Headington, Oxford

• Posterior Probability Enhanced Integration
  Modifies the class assignment after the HMRF-EM
  algorithm through adjustment of the posterior
  probability using Bayes Theorem. The
  posterior probability of each pixel is updated
  with corresponding spectral energy value through
  a simple addition. The probability to be added or
  deleted is equally spaced between the extents of
  energy scale.
• Results
• Figure 4 (a) HIFU liver B-mode image. (b)
  Cavitation region is drawn by hand. (c) Rayleigh
  HMRF-EM alone. (d) Intensity Enhanced
  Integration. (e) Prior Probability Enhanced
  Integration. (f) Posterior Probability Enhanced
  Integration.

• Introduction
• The high power intensities in HIFU often result
  in bubble production, either through cavitation
  or boiling, which are believed to be a primary
  contributor to tissue necrosis. Some HIFU
  protocols rely on the evidence of cavitation as a
  strong indicator of tissue lesions. Cavitation is
  associated with hyperechoic regions (bright up)
  in the image. The use of ultrasound (US)
  visualization for the guidance and monitoring of
  HIFU therapies most often relies on the
  appearance of a bright hyper-echoic region in the
  US B-mode image. Therefore reliable detection and
  the availability of a history of the occurrence
  and properties of cavitation is important for
  monitoring treatment.
• An underlying probabilistic method has been
  presented previously which automatically
  identifies hyperechoic regions spatially and
  temporally. However a number of difficulties are
  present in identifying cavitation regions.
• Variation in attenuation from bubble formation.
• Uncertainty near bright boudaries.
• Hyperechoic regions may appear for reasons
• other than cavitation tissue interface
• Therefore, we introduced an Active Cavitation
  Detection through spectral response of r.f.
  signal to attain high spatial resolution. This
  technique provides spectral information at a
  spatial resolution of about 1mm/pixel through
  using ARMA modeling and is measured in term of
  power spectrum density in high frequency
  emissions (over 10 MHz).

Figure 2 The B-mode image after the
deconvolution process
The average power spectrum density of region 1
and 2 was demonstrated for the pixels shown to
have high spectral energy (red yellow). It is
clear that the bright region shows little change
near the HIFU focus whereas significant change is
observed in region 2.
Figure 3 (a) The plot of power spectrum density
after deconvolution process in two regions 1 and
2 region 2 cavitation region and region1
tissue interface (Boundary). (b)Magnified view in
high frequencies.
Automatic Statistical Model (Intensity based)
Based on the Bayesian framework Uses Hidden
Markov Random Field with Expectation
Maximisation Maximise the posterior probability
of the associated class labels, x, to
intensity y
Details of Model 1. Noise model For each
pixel i, is noise probability
density function. Approximated by Rayleigh
p.d.f.. 2. Spatial models Assigns a prior
probability , using the Gibbs distribution
in Markov Random Field to encourages neighbouring
pixels to have the same class labels. The
summation is over the neighbouring eight pixels
and d is the Dirac delta function Iterative
solution using Expectation Maximisation
Initial estimation of noise model parameters
through Otsus Method.
Figure 1 (a) HIFU liver B-mode image. (b) Energy
color map showing the region around the focus
with magnified view too.
Note The circled region indicated in Figure 5(a)
persists but there is no strong energy response
in high frequency associating with that region.
Moreover, this region cannot be indicated as a
tissue lesion, because it disappeared after a
few seconds later in the subsequent image. It is
possible that it indicates boiling.
Aim To determine and develop displays which can
be used during treatment to assist the clinical
delivery of HIFU for example measures of the
cavitation region such as power intensity
distribution and persistence of bubbles, and
other factors of significance in heating and
lesion formation (possibly boiling
effect). Spectral estimation will be incorporated
with B-mode processing to enhance the overall
reliability of cavitation detection, together
with the comparison of three different
integration methods.
Conclusion and Further Work This work
investigated robust cavitation detection through
analysing both the spectral and intensity
properties of the r.f. (A-line) data underlying
B-mode images. Monitoring high frequency
emissions from cavitation provides both spatial
and spectral resolution through ARMA modelling of
r.f data. The information extracted here can be
used to improve the overall reliability of
cavitation segmentation and is capable
potentially of distinguishing between cavitation
and boiling. Further work is required to
investigate the boiling bubble effect, possibly
using listening hydrophones or transducers over a
significant frequency band to detect sub and
super harmonics at the same time as imaging.
ARMA Model (Spectral based)
Parameters of AR and MA model
Innovation(error term) is a white noise
process with zero mean variance Base on
. Parameters may be
calculated using Yule-Walker equation (yielding
p1 equations) and solve it as a
matrix. Integration methods 1. Intensity
enhanced Spectral information modifies the B
mode image. Pixels with energy above a certain
threshold in the spectral analysis have
intensity raised to maximum in the B-mode
image. 2. Prior Probability Enhanced
Integration Modifies the initial value of
the prior probability of each pixel class in
according to the spectral energy distribution of
the pixel. Pixels of highest energy having
unity prior probability, and a linear scale
down to those with zero high energy content
with probability 0.
Method Monitoring Change Deconvolution
Anticipated high spectral energies are observed
in the hyperechoic region near the focus, but
also near the interface. To separate out static
and dynamic effects, the heating response, which
we assume to be a bubble response, can be
derived using a deconvolution process in r.f data
processing. where b is the bubble response.
f is the current image, i, and, n is the sample
number of each the r.f. line. is the
original image, before heating.

• References
• J. McLaughlan, I. Riverns and G ter Haar,
  Cavitation detection in ex vivo bovine liver
  tissue exposed to high intensity focused
  ultrasound(HIFU), Biomedical Imaging From Nano
  to Macro. ISBI 2007. 4th IEEE International
  Symposium. pp 1124-1127, April 2007
• C.-C. Coussios. C.H. Farny, G. ter Haar and R.A.
  Roy , Role of Acoustic Cavitation in the
  Delivery and Monitoring of Cancer Treatment by
  High Intensity Focussed Ultrasound (HIFU),
  International Journal of Hyperthermia (2007). vol
  23, no 2, pp 105-120, 2007.
• K. Hynynen, The threshold for thermally
  significant cavitation in dogs thigh muscle in
  vivo, Ultrasound Med. Biol. vol 17, 157-169,1991