| Summary: | Ultrasound modulated optical tomography is a hybrid imaging modality with numerous potential clinical applications. In this work we develop, validate, and assess the accuracy of a number of forward models which describe the effect. We subsequently derive an inversion procedure which reconstructs images of the optical absorption coefficient in a turbid media from measurements of the optical field autocorrelation function made on the boundary. We begin with the development of a reference forward model which is accelerated by execution on the parallel architecture of modern graphics processing units. The model is validated against analytical and numerical results from the literature. The acceleration of the model results in improvements in performance of between one and two orders of magnitude when compared with a standard implementation. Whilst accurate, the reference model is not suited for use in an image reconstruction algorithm. As such, a number of alternative models are derived based upon spherical harmonic expansions of a correlation transport equation. In particular, a third order simplified spherical harmonic approximation is developed which provides a peak improvement in accuracy of over 50% relative to standard diffusion-style approximations. Each of the models is solved by the finite- element method, and validated against the reference Monte-Carlo simulation. Finally, we develop an image reconstruction procedure for ultrasound modulated optical tomography which employs a finite-element implementation of the proposed forward models. As part of the derivation we investigate the form of the correlation measurement density functions which describe the sensitivity of the technique to perturbations in the optical parameters of the medium. We demonstrate the ability to reconstruct an image of the optical absorption coefficient in a turbid medium from noisy measurements of the field autocorrelation function made on the boundary of the domain. Reconstructions employing data corrupted by 1% Gaussian noise achieve accuracy of circa 80% in the region of peak optical sensitivity, and maintain spatial resolution equivalent to the dimensions of the focused acoustic field probing the domain. Away from the sensitive region of the optical axis, regularisation forces the imaging resolution towards that of conventional diffuse-optical tomography.
|
|---|
