Fast Full Waveform Inversion Using Reflection Image and Anatomical Prior for Dual-Linear-Array Ultrasound Computed Tomography

Ultrasound computed tomography (USCT) based on full waveform inversion (FWI) has the capability to reconstruct high-resolution images. However, it often encounters challenges such as intensive iterations and being trapped in local minima caused by inaccurate initial guesses. To address these issues, we propose to build an accurate initial model while limiting the update gradient for each iteration of FWI, thereby mitigating cycle skipping effects and reducing computational effort. Compare to conventional approach, our approach integrates two imaging methods, reflection and transmission, while incorporating anatomical prior information. Firstly, we utilize the accurate tissue contour structure obtained from the reflection imaging results to minimize the computational domain, thereby reducing computational costs. Subsequently, by combining this contour information with anatomical prior knowledge, we construct a more precise initial model using the reflection imaging results. This approach reduces the number of iterations required in the full waveform inversion process, thereby decreasing the imaging time. The numerical simulation results demonstrate that our method, utilizing a high-precision initial acoustic model, yields significantly lower mean square error (MSE) in the reconstruction results compared to the conventional method, which employs a uniform background initial model, and our method exhibits faster convergence as the number of iterations increases. In terms of structural similarity, the results of our method are significantly higher than those of conventional methods, and as the number of iterations increases, the structural similarity increases faster.