A Skull-Adaptive Framework for AI-Based 3D Transcranial Focused Ultrasound Simulation

TL;DR

This work tackles the bottleneck of expensive subject-specific tFUS acoustic simulations by introducing TFUScapes, a large-scale dataset of full-wave 3D simulations across $125$ skull anatomies and $2{,}500$ configurations at $f_0=500$ kHz, generated with a GPU-accelerated k-Wave pipeline. It couples TFUScapes with DeepTFUS, a transducer-aware 3D U-Net that fuses Fourier-encoded transducer embeddings via dynamic convolutions, FiLM, and cross-attention to predict normalized pressure fields from pseudo-CT volumes and transducer geometry. Through extensive ablations, the authors show that a weighted loss and transducer-conditioning components notably improve focal localization and pressure-field fidelity, with the full DeepTFUS model delivering the best focal-position accuracy ($2.89\pm2.14$ mm) and strong max-pressure estimation while offering substantial speed-ups over solver-based techniques. The publicly released TFUScapes dataset aims to accelerate reproducibility and development of data-driven acoustic surrogates in neurotechnology, potentially enabling faster planning and optimization for tFUS interventions, though clinical validation and broader skull/transducer diversity remain for future work.

Abstract

Transcranial focused ultrasound (tFUS) is an emerging modality for non-invasive brain stimulation and therapeutic intervention, offering millimeter-scale spatial precision and the ability to target deep brain structures. However, the heterogeneous and anisotropic nature of the human skull introduces significant distortions to the propagating ultrasound wavefront, which require time-consuming patient-specific planning and corrections using numerical solvers for accurate targeting. To enable data-driven approaches in this domain, we introduce TFUScapes, the first large-scale, high-resolution dataset of tFUS simulations through anatomically realistic human skulls derived from T1-weighted MRI images. We have developed a scalable simulation engine pipeline using the k-Wave pseudo-spectral solver, where each simulation returns a steady-state pressure field generated by a focused ultrasound transducer placed at realistic scalp locations. In addition to the dataset, we present DeepTFUS, a deep learning model that estimates normalized pressure fields directly from input 3D CT volumes and transducer position. The model extends a U-Net backbone with transducer-aware conditioning, incorporating Fourier-encoded position embeddings and MLP layers to create global transducer embeddings. These embeddings are fused with U-Net encoder features via feature-wise modulation, dynamic convolutions, and cross-attention mechanisms. The model is trained using a combination of spatially weighted and gradient-sensitive loss functions, enabling it to approximate high-fidelity wavefields. The TFUScapes dataset is publicly released to accelerate research at the intersection of computational acoustics, neurotechnology, and deep learning. The project page is available at https://github.com/CAMMA-public/TFUScapes.

Figures (4)

• Figure 1: Overview of the TFUScapes dataset generation pipeline for creating subject-specific transcranial focused ultrasound (tFUS) pressure fields using MRI-derived pseudo-CTs and GPU-accelerated k-Wave simulations.
• Figure 2: Overview of the proposed transducer-aware 3D U-Net framework, DeepTFUS, for predicting tFUS pressure fields from pseudo-CT anatomy and transducer geometry.
• Figure 3: Qualitative results comparing different loss functions.
• Figure 4: Qualitative results of different architecture choices.