Abstract

Background.  Ultrasound is an attractive choice for noninvasive thermometry to enhance tumor treatment using hyperthermia.  Our theoretical model [1] predicted and in-vitro experiments [2] verified that the change in backscattered ultrasonic energy (CBE) is monotonic (approximately 0.2~0.3dB/oC) with temperature in the hyperthermia range, motivating the usage of CBE for ultrasonic thermometry.  One limitation in measuring temperature-dependent CBE from ultrasound images is apparent motion in the images due to change of the speed of sound and motion of the tissue.  Previously, a block-matching motion-tracking method was used to compensate rigid motion, but was only successful for small regions [2].
Methods.  We have developed an algorithm for estimating and compensating non-rigid motion over large 2D or 3D regions.  The motion field was modeled to vary linearly over the region of interest and was estimated by maximizing the cross-correlation between the reference and subsequent images using optimization functions in MATLAB®.  Factors affecting performance of the algorithm were studied using simulation of images for multiple scatterers [3]. Images before and after motion were simulated by transforming scatterer locations.
Results.  Our algorithm was successfully applied to images from in-vitro and in-vivo heating experiments, with significant improvement in performance over previous results based on qualitative assessment.  To study the performance quantitatively, 2D images were simulated with various signal-to-noise ratio (SNR) and types and ranges of motion, including expansion and compression to study image decorrelation.  CBE due to motion was different between axial and lateral directions and was about 2.5dB/0.1mm translation and 1dB per ±1% expansion axially and 4.5dB/mm and 0.35dB per ±1% laterally.  Error in motion estimation decreased with SNR with minimal impact on CBE and increased with decorrelation leading to 0.1~0.2dB additional CBE.  In addition to inducing estimation error and thus CBE, decorrelation itself caused erroneous CBE as much as 0.3~0.4 dB for 6% lateral compression and 1% axial compression.
Conclusions.  CBE induced by motion critically limits temperature imaging and thus necessitates accurate motion estimation and compensation.  Our current algorithm works effectively with motion studied in simulation and encountered in heating experiments.  Future work will include incorporation of nonlinear motion field, estimation of motion in the presence of temperature-dependent CBE, reducing effects of decorrelation, and motion compensation on temperature imaging during clinical hyperthermia.