International Workshop on Medical Ultrasound Tomography

Source selection for ultrasound waveform tomography: real data case from USCT Data Challenge 2017

Nov 2, 2017, 11:00 AM

20m

Lecture Hall

Oral Main Track Session 6: System design

Speaker

Naiara Korta Martiartu (ETH Zurich)

Description

Waveform inversion for Ultrasound Computed Tomography (USCT) is an emerging high-resolution imaging technique for breast cancer screening. Despite its potential, the involved computational burden is challenging. This mainly depends on the total number of wave propagation simulations solved during the inversion procedure. Thus, the computational cost can be reduced by only selecting the emitting transducers that best resolve the information about the breast tissue and avoid redundant experiments.

(4) Discussion and Conclusion

We applied sequential optimal experimental design to optimize the transducer configuration of USCT scanning devices. With the aim of reducing the computational cost of the inverse problem, we provided a systematic framework to avoid collecting redundant information about the model and to quantify the information gain by adding more emitters. Furthermore, the method can be extended to other design parameters. In particular, it can be used in combination with source encoding strategies.

(2) Material and Methods

We identify the optimal emitting transducers using Sequential Optimal Experimental Design (SOED). SOED is a powerful tool to iteratively add the most informative transducer location or remove redundant experiments, respectively. This approach simultaneously provides optimized transducer configurations and a cost-benefit curve that quantifies the information gain versus the computational cost.

We measure the quality of different configurations quantifying the expected model uncertainties post-reconstruction. Using a Bayesian approach, this can be related to the properties of the posterior covariance which in general can be computed using the Hessian operator. For waveform inversion, we use the Gauss-Newton approximation of the Hessian.

(3) Results

We use real data provided by CSIC as part of the USCT Data Challenge 2017. The dataset consists of 64768 A-Scans collected by a moving array of 16 transducers emitting with a dominant frequency of 3.5 MHz. To select the sources to be used in waveform tomography, we first apply 2D straight rays approximation mainly for two reasons:

(1) To reconstruct the initial sound speed model for waveform inversion.

(2) To gain intuition about the most relevant emitting transducers. For such high frequencies, the first Fresnel zone of the sensitivity kernels used in the Gauss-Newton approximation of the Hessian is quite narrow. Thus, similar patterns are expected when applying SOED for waveform inversion. This will be verified by the work in progress.