Transcranial ultrasound imaging using pulse-echo ultrasound and deep learning: A numerical study

Abstract

Phase aberration caused by skulls is a main challenge in transcranial ultrasound imaging for adults. Aberration could be corrected if the skull profile (i.e., thickness distribution) and speed of sound (SOS) are known. We previously designed a deep learning (DL) model to estimate the skull profile and SOS using pulse-echo ultrasound signals. This study’s objective is to develop strategies to improve the estimation and evaluate the effectiveness of aberration correction in transcranial ultrasound imaging. Acoustic simulations were performed using k-Wave in this numerical study. The following strategies were used to improve estimation: (1) A phased array was used instead of a single-element transducer; (2) Channel radiofrequency data were used instead of beamformed data as the DL model input; (3) A DL model was developed to incorporate physics into architecture design and model training. Compared with previously reported results, these strategies improved the correlation coefficient between the estimated and ground-truth values from 0.82 to 0.94 for SOS, and from 0.98 to 0.99 for thickness. Simulated transcranial images of point targets with phase correction using the estimated SOS and thickness values showed significantly reduced artifacts than those without correction. The results demonstrate feasibility of the proposed approach for transcranial ultrasound imaging.