Session: 07-05-01: Dynamics and Control of Biomechanical Systems

Paper Number: 173922

Computational Mechanics of Transcranial Ultrasonic Stimulation 

Transcranial Ultrasound Stimulation (TUS) is a non-invasive therapeutic technique that delivers focused ultrasound waves to targeted regions of the brain. As the wave propagates through complex brain structures, it undergoes attenuation, refraction, and reflection, which, in turn, incurs challenges in precise focal control. These challenges are further exacerbated by the heterogeneous and anisotropic nature of brain tissue, potentially leading to unintended stimulation or other effects in off-target regions. Accurate numerical simulations are therefore essential for predicting these effects and minimizing risks to patients. Additionally, such simulations can inform the optimal design and arrangement of transducers for enhanced focusing accuracy.

In this presentation, we introduce a computational mechanics framework for simulating low-intensity TUS under small deformations assumption, with applications in neuromodulation. Ultrasonic waves are modeled as time-harmonic elastic waves in the frequency domain, with brain tissue simultaneously characterized as heterogeneous, anisotropic, and viscoelastic. We also adopt a phase conjugation method, which represents the frequency-domain analogue of time reversal method, for optimal focusing. We will also discuss some nuances of high precision focusing of ultrasonic waves and TUS. We determine the amplitude and phase distribution of the input loading required to focus energy at a prescribed target. The impact of tissue heterogeneity and anisotropy on off-target effects is quantitatively investigated in this research. Our results conclusively elaborates that heterogeneities significantly disturb ultrasonic wave focusing, and unless precisely accounted for through in-silico models, introduce potentially significant errors in efficacy of such therapies.

Moreover, we discuss issues in predicting high frequency pressure waves or modeling shear waves in brain tissue, and how, using traditional methods, capturing the ensuing small wavelengths proves to be a computationally intractable problem. We will then delineate that diffusion models  as probabilistic machine learning frameworks enable addressing such problems rendered in TUS applications. We demonstrate that diffusion models can learns the map from brain properties to steady state waves, essentially learning the solution operator for Helmholtz equations. Our approach benefits from the typical advantages of ML frameworks such as being able to serve as a Blackbox that quickly outputs results, once trained and validated. Additionally, we show how diffusion models enable both predicting energy deposited deep inside the brain, as well as allowing for a robust pathway for quantifying uncertainties incurred by patient specificity of the mechanical properties of brain. We benchmark diffusion model results with recent developments in neural operators and show that it can conclusively outperform other ML techniques in high frequencies.

Presenting Author: Hossein Salahshoor N/A

Presenting Author Biography: Professor (Amir)Hossein Salahshoor is an Assistant Professor in the departments of Civil and Environmental Engineering, and Mechanical Engineering and Material Science at Duke. Before that, he conducted his postdoctoral studies at Caltech. Prior to that he obtained his Ph.D in Aerospace Engineering from Georgia Tech, along with an MS in Mathematics. His research interests broadly lie at the intersection of mechanics of materials and structures, computational and data science, biology, and applied mathematics.

Authors:

Hossein Salahshoor N/A