Multi Taper Coherence Measurements for Characterizing Scatterer Number Density
N Rubert*, T Varghese, University of Wisconsin, Madison, WISU-E-I-101 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: Scatterer number density (SND) per resolution cell has been investigated by many authors as a useful tool for tissue classification in diagnostic ultrasound. SND has been shown to be capable of distinguishing between benign and malignant tumors in the breast, and of determining the degree of lens hardening in cataract formation. Low SND results in coherent scattering, which may be measured using the generalized spectrum (GS). The GS has been measured in ultrasound by correlating windowed FFT's of ultrasound radiofrequency (RF) data, where the data windows have been chosen ad hoc. We seek to improve GS calculations for distinguishing between tissue containing different number densities using Thomson's multi-taper method.
Methods: Three tissue-mimicking (TM) phantoms consisting of glass beads and graphite powder suspended in agar were constructed. Scattering was mainly due to the glass beads and attenuation was mainly due to the graphite powder. The phantoms had number densities of 2, 10, and 400 glass beads per mm³. RF data was recorded using a 9L4 and 18L6 transducer from a Siemens S2000 system. In this study we compare two different methods of calculating spectral coherence. In one method we use a single taper and convolutionally smooth the GS. In the second method we compute the GS using Thomson's multi-taper (MT) method. We show that the MT method outperforms diagonally smoothed single-taper spectral estimates for distinguishing between the different phantoms.
Results: We find that the MT method results in a greater separation between the mean coherence values with a lower standard deviation than single-taper calculation methods, as the SND is varied.
Conclusion: When SND is used to distinguish between different tissues, differences will be better captured if coherence is calculated with Thomson's multi-taper method rather than a single taper.
This work was supported by NIH-NCI grants R01CA112192-06 and T32 CA09206-32.