Model reduction in acoustic inversion by artificial neural network
This work addresses computational efficiency and image quality for ultrasound tomography applications, representing an incremental improvement in domain-specific methods.
The paper tackles the problem of slow computations in ultrasound tomography due to accurate forward models by proposing a neural network to compensate for errors from approximate models, resulting in significantly improved image reconstruction quality with small training datasets compared to common inversion algorithms.
In ultrasound tomography, the speed of sound inside an object is estimated based on acoustic measurements carried out by sensors surrounding the object. An accurate forward model is a prominent factor for high-quality image reconstruction, but it can make computations far too time-consuming in many applications. Using approximate forward models, it is possible to speed up the computations, but the quality of the reconstruction may have to be compromised. In this paper, a neural network -based approach is proposed, that can compensate for modeling errors caused by the approximate forward models. The approach is tested with various different imaging scenarios in a simulated two-dimensional domain. The results show that with fairly small training datasets, the proposed approach can be utilized to approximate the modelling errors, and to significantly improve the image reconstruction quality in ultrasound tomography, compared to commonly used inversion algorithms.