Acoustic neural network thermometry in monitoring of thermal therapies.

Abstract

The potential benefits of monitoring tissue temperatures and lesioning by “acoustic” methods are evident. Such methods are typically challenging, and limited to interrogating (acoustically) only one or two physical processes during dosing (e.g., sound speed and thermal expansion), which may narrow valid temperature ranges. Due to the complexity of tissue behavior and acoustic interactions during heating, a “gestalt” of acoustic parameters was enabled to be “learned” during dosing by a recurrent neural network (RNN). Using this method, real‐time three‐dimensional (3‐D) acoustic thermometry was demonstrated and (from RNN coefficients) insights were gained into specific acoustic parameter contributions accompanying temperature changes during and after dosing. The RNN method is described for two applications. First, for radiofrequency ablation on excised bovine livers, coregistered B‐mode and ARFI elasticity images were acquired. RNN thermometry images were then calculated from rf data, thermal doses calculated, and the necrosis region was then determined, and shown in good agreement with elasticity lesion images. Secondly, HIFU was used to ablate excised bovine liver regions, and the RNN temperatures in tissues were shown to reflect projected power depositions of the HIFU transducer. These results support RNN thermometry as potentially suitable for some thermal therapies. [This research sponsored by U.S. DARPA Contract No. HR0011‐08‐3‐0004.]