Reduction of Image Noise in Elastography

1
Reduction of Image Noise in Elastography

• By Ignacio Cespedes Jonathan Ophir
• Presented by Harish Krishnaswamy
• Partner Parker Wilson
• Mentors Emad Boctor, Dr. Russell Taylor

2
Paper Selection

• Basis of Ophirs Algorithm.
• Core algorithm in retrieving and refining strain
  data using real Analytical Signals.
• We use this in addition to Lorenzs Algorithm to
  develop our Elastogram.

3
Introduction

• Elastography is a method of imaging the elastic
  properties of compliant tissues which produces
  gray scale elasticity images called Elastograms.
• The method estimates local strain directly from
  estimates of one dimensional changes in small
  tissue elements.
• Using A-lines obtained pre and post target
  compression.

4
Problem

• However, Elastograms of Phantoms with homogenous
  elastic properties exhibit a noisy appearance.
• Mainly due to nonstationary relation between pre
  and post compression signals.
• How to reduce the strain modulation artifact
  while maintaining improved spatial resolution?

5
Strategy

• Two methods to reduce the strain modulation
  artifact
• Reduce the signal amplitude swings within the
  observation windows by logarithmically
  compressing the rf signal.
• Reduce the strain modulation by temporally
  stretching the signal obtained after compression.

6
Elastographic Noise

• Factors that contribute to the noisy appearance
  of Elastograms.
• Elasticity of Phantom is not globally
  homogeneous.
• Errors in time shift correlation estimations
  result in temporal uncertainty that causes random
  noise in local strain estimates.
• All of which contribute to degraded
  cross-correlation.

7
Strain Modulation Artifact

• Estimates of strain are obtained by measuring the
  displacement between pre and post target
  compression.
• Consecutive observation window pairs are taken
  with a predetermined spacing of ?T.
• Thus, after target compression the spacing
  between consecutive window pairs is reduced by
  amount ?t(i)-?t(i-1)

8
Ophir et al.
Two consecutive Non overlapping windows. Note the
reduced separation between pre and post target
compression A - Lines
9
Concepts

• Local Longitudinal Strain is given by
• s(i) ?t(i) - ?t(i-1)/ ?T
• Due to Deformation of rf signal after target
  compression we need to consider the following
  concepts
• Interwindow Compression Spacing between
  consecutive window pairs before and after
  compression contains elasticity information
  (Virtual Spring).
• Intrawindow Compression The signal from
  compressed tissue is distorted such that it
  resembles a time scared version of the pre-target
  compression signal.

10
Impaired Cross-Correlation
Ophir et al.
11
Theory

• Logarithmic Compressing of rf Amplitude
• Temporal stretching

Original Signal
Temporal Stretching
Compressed Signal
12
Experiment

• Algorithms were tested on simulated data
  generated by a computer program.
• Simulated program contained the following
• An ultra sound transducer
• A model for the compliant scattering medium
• Simulated A-lines were constructed by adding up
  the impulse responses of the transducer at the
  locations of all scatters in a defined region of
  interest.

13
Ophir et al.
Equivalent spring model for columnar tissue
element as used in the simulation program.
Springs represent segments of the tissue column
with different Youngs moduli.
14
Theory Cont.

• Logarithmic amplitude compression is applied to
  digitized rf signal
• One Bit Quantization
• If y(n) 0 then y(n) 1.0
• If y(n) 0 then y(n) -1.0
• Temporal stretch was applied using a linear
  interpolation Algorithm.

15
Analysis

• Quantification of Image Quality was conducted
  using mean-to-standard deviation ratio (MDSR)
• Where µs and ss are respectively the mean and
  standard deviation of the strain in a region of
  uniform elasticity.

16
Results
Ophir et al.
17
Results Cont.
MDSR Values
Ophir et al.

• MDSR is lowest without log compression,
  increasing rapidly for log compression strength
  values up to CS100, then decreasing slowly for
  larger log compression strengths.
• One bit transformation yields MSDR that is
  inferior to the log compression case.
• This is a result of the reduced distortion of
  the pre and post compression signal.

18
Conclusion

• Logarithmic amplitude compression achieves the
  desired reduction of the variability of the rf
  envelope.
• Improvement in MSDR using logarithmic amplitude
  compression is achieved without sacrifice in
  spatial resolution.
• Note Resolution worsens when temporal
  stretching is applied.