Flawed transducer detection using random sample consensus for ultrasound tomography

In this paper, we present a random sample consensus (RANSAC) based ultrasound travel-time tomography method. Conventionally, all the time-of-flight (TOF) data between each two transducers are used to estimate the sound speed distribution. However, failing to identify the inaccurate TOF data (outliers) due to flawed transducers would reduce the accuracy of the estimated sound speed distribution. In our proposed approach, a small subset of TOF data were first randomly selected from the original TOF data, and then applied to the tomography algorithm to estimate a rough sound speed distribution. The rest of the TOFs data were applied to the rough distribution and the goodness of fit was calculated. If most of the data fitted well in the estimated distribution, then all the well-fitted data (including the subset) was used to estimate a final sound speed distribution. Otherwise, outliers were expected in the subset and a new subset of the TOFs data would be randomly selected again. This repeated until most of the data fitted well in the estimated distribution. Simulation results showed that our method could effectively detect and eliminate outliers and increase the accuracy for estimating the sound speed distribution.