Learning-Based Approach for Real-Time Versatile Ultrasound Computed Tomography

Ultrasound computed tomography, implemented on the basis of circular transducer array in general, is known to provide quantitative characteristics of imaging objects. However, existing UCT solutions are not popular in clinical applications due to their usage restrictions and long reconstruction time. In this paper, we propose a versatile UCT system employing two facing transducer arrays to gather and analyze bidirectional reflected and traversed waves. A Neural Network (NN) approach is incorporated to reduce the acquisition time for real-time image reconstruction. To gradually supplement the details of lesions, a refinement network structure is proposed and the quality of reconstructed image exceeds the quality of conventional systems when judged by diverse test metrics.