Real-time Simulation of Phased Multi-element Transducers Using AI

Abstract

Objectives: We propose an AI model that predicts the acoustic pressure distribution in real-time based on the phase information of phased array transducers.

Methods: The k-Wave Matlab toolbox was utilized to generate acoustic simulation data, which served as the ground-truth for training the AI model. The model was structured as an encoder-decoder network based on CNN and Transformer with a U-shaped design. It takes geometric information of the transducer and phase information of each element as input and yields the simulation result. A 500 kHz bowl-shaped transducer with 4 channels was employed, which is widely used in tFUS neuromodulation.

Results: Along with the respective processing times for each method, we evaluated the accuracy of our AI model by comparing the peak pressure ratio and the dice similarity coefficient (DSC) of full-width at half-maximum (FWHM) between the AI inference and result of k-Wave simulation. In one simulation, while k-Wave took 8 s, the AI model completed the inference in just 0.025 s. For 300 cases where the phase was randomly assigned, the average peak pressure error was 5.6%, and the average DSC was 88.94% in free water. The image below provides an example comparing the steered focus according to phase variations.

Conclusions: We introduce an AI model for real-time prediction of acoustic pressure distribution based on phase variations. Numerical evaluations show comparable results to k-Wave simulation with reduced processing time, facilitating real-time monitoring of steered focal positions. Future studies will explore applying transcranial acoustic pressure on skull models with AI model.