tFUSFormer: Physics-Guided Super-Resolution Transformer for Simulation of Transcranial Focused Ultrasound Propagation in Brain Stimulation

Abstract

Transcranial focused ultrasound (tFUS) has 1 emerged as a new mode of non-invasive brain stimulation 2 (NIBS), with its exquisite spatial precision and capacity to 3 reach the deep regions of the brain. The placement of the 4 acoustic focus onto the desired part of the brain is critical 5 for successful tFUS procedures; however, acoustic wave 6 propagation is severely affected by the skull, distorting the 7 focal location/shape and the pressure level. High-resolution 8 (HR) numerical simulation allows for monitoring of acoustic 9 pressure within the skull but with a considerable compu-10 tational burden. To address this challenge, we employed 11 a 4x super-resolution (SR) Swin Transformer method to 12 improve the precision of estimating tFUS acoustic pressure 13 field, targeting operator-defined brain areas. The training 14 datasets were obtained through numerical simulations at 15 both ultra-low (2.0 mm) and high (0.5 mm) resolutions, con-16 ducted on in vivo CT images of 12 human skulls. Our mul-17 tivariable datasets, which incorporate physical properties 18 of the acoustic pressure field, wave velocity, and skull CT 19 images, were utilized to train three-dimensional SR models. 20 We found that our method yielded 87.99±4.28% accuracy 21 in terms of focal volume conformity under foreseen skull 22 data, and accuracy of 82.32±5.83% for unforeseen skulls, 23 respectively. Moreover, a significant improvement of 99.4% 24 in computational efficiency compared to the traditional 0.5 25 mm HR numerical simulation was shown. The presented 26 technique, when adopted in guiding the placement of the 27 FUS transducer to engage specific brain targets, holds 28 great potential in enhancing the safety and effectiveness 29 of tFUS therapy.30