eduzhai > Applied Sciences > Engineering >

Training Variational Networks with Multi-Domain Simulations Speed-of-Sound Image Reconstruction

  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Speed-of-sound has been shown as a potential biomarker for breast cancerimaging, successfully differentiating malignant tumors from benign ones.Speed-of-sound images can be reconstructed from time-of-flight measurementsfrom ultrasound images acquired using conventional handheld ultrasoundtransducers. Variational Networks (VN) have recently been shown to be apotential learning-based approach for optimizing inverse problems in imagereconstruction. Despite earlier promising results, these methods however do notgeneralize well from simulated to acquired data, due to the domain shift. Inthis work, we present for the first time a VN solution for a pulse-echo SoSimage reconstruction problem using diverging waves with conventionaltransducers and single-sided tissue access. This is made possible byincorporating simulations with varying complexity into training. We use loopunrolling of gradient descent with momentum, with an exponentially weightedloss of outputs at each unrolled iteration in order to regularize training. Welearn norms as activation functions regularized to have smooth forms forrobustness to input distribution variations. We evaluate reconstruction qualityon ray-based and full-wave simulations as well as on tissue-mimicking phantomdata, in comparison to a classical iterative (L-BFGS) optimization of thisimage reconstruction problem. We show that the proposed regularizationtechniques combined with multi-source domain training yield substantialimprovements in the domain adaptation capabilities of VN, reducing median RMSEby 54 on a wave-based simulation dataset compared to the baseline VN. We alsoshow that on data acquired from a tissue-mimicking breast phantom the proposedVN provides improved reconstruction in 12 milliseconds.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...