A hierarchical framework for robust full-wave inversion in transcranial thermoacoustic tomography

Abstract

Skull-induced acoustic heterogeneity, multipath propagation, and waveform corruption severely degrade structural reconstruction in transcranial thermoacoustic imaging. Conventional methods such as time-reversal and standard full-wave inversion (FWI) typically assume clean measurements or weak heterogeneity, leading to compromised accuracy and stability under realistic conditions. To jointly address these coupled challenges, we propose a hierarchically integrated framework to reduce reconstruction errors, integrating robust principal component analysis for measurement-level signal purification, wave-equation-based FWI for modeling-level mismatch correction, and adaptive gradient modulation for optimization-level convergence stabilization. Numerical simulations and phantom experiments demonstrate that the proposed method achieves clearer boundary definition and more stable reconstructions compared to time-reversal and conventional FWI-based approaches, particularly under noisy and limited-view acquisitions. Quantitatively, noisy numerical simulations reduce the overall reconstruction root mean square error by approximately 8%, improve the structural similarity index by approximately 52%, and increase the peak signal-to-noise ratio by approximately 0.8 dB relative to conventional approaches under representative transcranial imaging scenarios. This physics-guided hierarchical design offers an effective and robust strategy for reliable structural imaging in transcranial thermoacoustic tomography.