Files and links (1)
Rights Statements InC/1.0 , Open Access
Journal article
A comparison of material classification techniques for ultrasound inverse imaging
The Journal of the Acoustical Society of America., Vol.111(1), pp.457-467
01/2002
Handle:
https://hdl.handle.net/2376/6011
Abstract
The conjugate gradient method with edge preserving regularization (CGEP) is applied to the ultrasoundinverse scattering problem for the early detection of breast tumors. To accelerate image reconstruction, several different pattern classification schemes are introduced into the CGEP algorithm. These classification techniques are compared for a full-sized, two-dimensional breast model. One of these techniques uses two parameters, the sound speed and attenuation, simultaneously to perform classification based on a Bayesian classifier and is called bivariate material classification (BMC). The other two techniques, presented in earlier work, are univariate material classification (UMC) and neural network (NN) classification. BMC is an extension of UMC, the latter using attenuation alone to perform classification, and NN classification uses a neural network. Both noiseless and noisy cases are considered. For the noiseless case, numerical simulations show that the CGEP–BMC method requires 40% fewer iterations than the CGEP method, and the CGEP–NN method requires 55% fewer. The CGEP–BMC and CGEP–NN methods yield more accurate reconstructions than the CGEP method. A quantitative comparison of the CGEP–BMC, CGEP–NN, and GN–UMC methods shows that the CGEP–BMC and CGEP–NN methods are more robust to noise than the GN–UMC method, while all three are similar in computational complexity.
Metrics
80 File views/ downloads
51 Record Views
Details
- Title
- A comparison of material classification techniques for ultrasound inverse imaging
- Creators
- Xiaodong Zhang (Author)Shira L. Broschat (Author)Patrick J. Flynn (Author)
- Publication Details
- The Journal of the Acoustical Society of America., Vol.111(1), pp.457-467
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Identifiers
- 99900502792301842
- Copyright
- In copyright ; openAccess ; http://rightsstatements.org/vocab/InC/1.0/ ; http://purl.org/eprint/accessRights/OpenAccess
- Language
- English
- Resource Type
- Journal article