Jul 2, 2012
Microwave methods tackle breast imaging
Breast imaging at microwave frequencies exploits the variation in dielectric properties between different breast tissue types to create an image. Importantly, differences in the dielectric properties of malignant and normal breast tissues have been reported, lending support to the use of this technique for cancer diagnoses.
Two approaches to microwave imaging are proposed: radar-based imaging, in which the breast is illuminated with ultrawideband pulses and reflection data are recorded; and microwave tomography, which uses additional receivers to record both reflection and transmission data. Radar images indicate the locations of strongly scattering objects, while microwave tomography creates a low-resolution map of the tissue's dielectric properties. Reconstruction of this map, however, requires solving of an ill-posed inverse scattering problem.
Researchers at the University of Calgary in Canada have come up with a new take on this challenge by integrating the two techniques. The idea is to reduce the ill-posedness of the problem by using prior information, in this case the radar data, to define the breast's internal structure. The radar data are used to segment the breast into three regions – a skin layer and two inhomogeneous regions dominated by adipose and fibroglandular tissues – prior to evaluation of each region's mean dielectric properties (Inverse Problems 28 075001).
"There is potential for the two techniques to work together synergistically," explained researcher Douglas Kurrant. "For example, the tomography map provides basic structural information about the breast, while radar images may be used to detect locations of interest for further examination. The low-resolution maps may be also used to improve radar-based imaging techniques, which typically assume an homogeneous breast composition."
The proposed technique required development of a series of algorithms. A contour sample evaluation algorithm uses the radar reflection data to identify the boundary locations, while a reconstruction model estimation algorithm estimates the locations of the three regions of interest. Finally, Kurrant and co-author Elise Fear developed a parameter estimation algorithm that uses microwave tomography data to determine the average dielectric properties for each region.
Kurrant and Fear evaluated the performance of their proposed algorithm using a series of increasingly complex three-component models, starting with a simple model in which the three regions were relatively homogeneous and well separated. The regions represented skin, adipose and fibroglandular tissues, and were assigned appropriate values of relative permittivity and conductivity.
Reflection data generated by scanning 40 equally spaced locations around the model were used to identify the three regions. The researchers then applied the parameter estimation algorithm to transmission and reflection data collected by moving a source and two sensors around 16 equally spaced locations.
Overall, the algorithm estimated the mean dielectric properties for skin and fibroglandular tissue with high accuracy. The adipose tissue region exhibited larger discrepancies, both in the error ratio (a quantitative measure of how accurately geometrical properties can be modelled) and the relative error of the dielectric parameters.
The researchers noted that adding noise to the model reduced the quality of the dielectric parameter estimates for the adipose region, although the relative error remained below 15% for a signal-to-noise ratio (SNR) of 20 dB. There was only a marginal increase in geometric errors as the SNR decreased, highlighting an advantage of integrating radar-based techniques with tomography.
They then assessed a second model in which the fibroglandular region was non-homogeneous and could not be well segregated from the adipose region. Following the above approach, the reconstructed profiles exhibited larger error ratios for adipose and fibroglandular regions (39.97% and 15.88%, respectively) than seen for the simpler model (12.8% and 6.01%). Despite an increase in relative errors, the algorithm provided reasonably accurate estimates of mean dielectric properties.
Finally, Kurrant and Fear examined two numerical breast models constructed from patient MR scans. After applying the contour sample evaluation algorithm to the reflection data, a general outline of the fibroglandular region could be extracted. However, this region had a far more complex shape than the previous models studied, and its contour could not be precisely delineated. Error ratios for the adipose and fibroglandular regions were 40.3% and 29.8% for one of the models, and 31.3% and 30.8% for the other.
The parameter estimation algorithm provided accurate estimates of dielectric properties for the skin and fibroglandular region; larger relative errors were seen for the adipose region. The authors note that despite the highly heterogeneous nature and complex shape of the fibroglandular region, and the presence of isolated fibroglandular scatterers within the adipose region, the results support the feasibility of simplifying the breast's internal structure to just three predominant tissue types.
The researchers concluded that their technique can efficiently provide low-resolution images, along with estimates of the average dielectric properties over regions dominated by the skin, adipose and glandular tissues. "We believe that the algorithm represents a practical improvement for microwave imaging," said Kurrant. "We are currently extending the algorithm to three dimensions, and plan to apply it to experimental data to demonstrate feasibility for practical problems in microwave breast imaging."
About the author
Tami Freeman is Editor of medicalphysicsweb.