To unlock the full capability of photoacoustic tomography as a quantitative, high resolution, molecular imaging modality, the problem of quantitative photoacoustic tomography must be solved. The aim in this is to extract clinically relevant functional information from photoacoustic images by finding the concentrations of the chromophores in the tissue. This is a challenging task due to the effect of the unknown but spatially and spectrally varying light fluence within the tissue. Many inversion schemes that include a model of the fluence have been proposed, but these have yet to make an impact in pre-clinical or clinical imaging. In this study, the statistical independence of the chromophore's distributions is proposed as a means of improving the robustness and hence the usefulness of the model-based inversion methods. This was achieved by minimising the mutual information between the estimated chromophore distributions in addition to the least squares data error within a gradient-based optimisation scheme. By applying the proposed inversion scheme to simulated multiwavelength photoacoustic images, it was shown that more accurate estimates for the concentrations of independent chromophores could be obtained in the presence of errors in the model parameters.
The ability to accurately quantify chromophore concentration from photoacoustic images would have a major impact on pre-clinical and clinical imaging. Recent years have seen significant advances in the theoretical understanding of quantitative photoacoustic imaging and in the development of model-based inversion strategies that overcome issues such as non-uniqueness and non-linearity. Nevertheless, their full in vivo implementation has not successfully been achieved, partially because experimental uncertainties complicate the transition. In this study, a sensitivity analysis is performed to assess the impact on accuracy of having uncertainty in critical experimental parameters such as scattering, beam diameter, beam position and calibration factor. This study was performed using two virtual phantoms, at one illumination and four optical wavelengths. The model-based inversion was applied in 3 variants - one just inverting for chromophores and two others further inverting for either a scaling factor or the scatterer concentration. The performance of these model-based inversions is also compared to linear unmixing strategies - with and without fluence correction. The results show that experimental uncertainties in a priori fixed parameters - especially calibration factor and scatterer concentration - significantly affect accuracy of model-based inversions and therefore measures to ameliorate this uncertainty should be considered. Including a scaling parameter in the inversion appears to improve quantification estimates. Furthermore, even with realistic levels of experimental uncertainty in model-based input parameters, they outperform linear unmixing approaches. If parameter uncertainty is large and has significant impact on accuracy, the parameter can be included as an unknown in model-based schemes.
Quantitative photoacoustic tomography involves the construction of a photoacoustic image from surface measurements of photoacoustic wave pulses and the recovery of the optical properties of the imaged region. This is a nonlinear, ill-posed inverse problem, for which model-based inversion techniques have been proposed. Here, the radiative transfer equation is used to model the light propagation, and the acoustic propagation and image reconstruction are included. In other words, the full quantitative inversion is tackled. Since Newton-based minimisations are impractical when dealing with three-dimensional images, an adjoint-assisted gradient-based inversion was used as a practical alternative to determining the optical coefficients.