R. M. Arthur, J. W. Trobaugh, W. L. Straube, J. Parry, Y. Guo and E. G. Moros, "Change in Ultrasonic Backscattered Energy for Temperature Imaging:  I. Simulation with Multiple Scatterers, II. Measurements from In Vivo Images" 31st International Symposium on Ultrasonic Imaging and Tissue Characterization, Arlington, Virginia, 24 – 26 May  2006.

Abstract

    Ultrasound is an attractive modality for non-invasive temperature imaging to enhance the ability to target tumor heating at therapeutic levels.  Previously, we measured monotonic changes in ultrasonic backscattered energy (CBE) in vitro in 2D and in 3D that matched changes we predicted for certain sub-wavelength scatterers.  Here we consider: 1) measurement of CBE in 2D in vivo, 2) simulation of CBE from multiple scatterers, and 3) estimation of temperature from CBE in simulated images.

    We measured CBE in living normal murine tissue and in implanted tumors (HT29 colon cancer line) on nude-mouse preparations.  Measurements were made in degassed water heated homogeneously.  Four mice, one with an implanted tumor, were anesthetized with Ketamine Xylazine and heated.  Temperature was measured with a thermistor at the hind limb contralateral to the one imaged.  Imaging was done with a Terason 2000 (Teratech Corp., Burlington, MA), laptop-based, phased-array system.  The imaging system used a 7-MHz linear probe (model 10L5) focused at 2 cm, the center of the mouse leg.  Images were taken in 0.5^oC steps from 37.0 to 45.0^oC.  For image regions within each preparation, non-rigid motion compensation was applied using cross-correlation of RF signals at adjacent temperatures.  Envelopes of motion-compensated image regions were found with the Hilbert transform then smoothed with a 3x3 running average filter. Backscattered energy at each pixel was referred to the value at 37^oC to find CBE.  CBE was nearly monotonic with temperature.  BE differed by 5-6 dB at 45^oC from its value at 37^oC.

    Theoretical results for a single scatterer showed that backscattered energy increased or decreased monotonically, depending on the lipid or aqueous nature of the scatterer.  To extend our theory to a more realistic tissue composition, we have developed methods for simulating ultrasonic images of thousands of randomly distributed scatterers.  In the simulations, the imaging system was described by its point-spread function.  The tissue medium was represented by discrete aqueous and lipid scatterers.  Images were simulated to represent temperatures from 37 to 50^oC by changing the scatterer amplitudes according to curves predicted previously for single scatterers.  CBE was computed for each image pixel, referenced to the initial image. To characterize CBE for a region, the means of the positive- and negative-changing pixels and the standard deviation of all pixels were computed.  CBE showed the same monotonic increase and decrease as in experimental results and covered ranges similar to both prediction and experiment. Subsequent simulations included additive noise and showed striking agreement with experimental CBE measurements, replicating both an initial jump and noise throughout the range. These results support the use of CBE for noninvasive temperature estimation, showing that our model for the temperature dependence of CBE can be successfully applied to measurements from multiple scatterers.  These simulation methods also provide a means for exploring limits on temperature accuracy and spatial resolution with varying imaging systems and tissue types.

    As an example of methods that could be used for calibration and estimation, we have extended our image simulation methods to the initial development of calibration curves and use of those curves for estimating temperature from CBE.  Calibration curves were generated by fitting the average CBE from multiple simulations with a polynomial.  This polynomial represents the average standard deviation of the CBE for images of a simulated population of lipid and aqueous scatterers with added noise. Estimates of temperature were then generated from that same population using the calibration curve.  Results showed an error of approximately ± 1^oC for these images of 1 cm² or approximately 0.3 cm³ (based on a 3 mm elevation ultrasonic beam width).  This error for a small volume with typical measurement noise levels suggests 0.5^oC or better accuracy may be attainable in 1 cm³ volumes with noise reduction techniques.

Support:  R21-CA90531, R01-CA107558 and the Wilkinson Trust at Washington University, St. Louis.