Optoacoustic tomography can generate high-resolution optical images of biological samples in vivo at depths of several millimeters to centimeters. The technique is based on illuminating the sample with nanosecond laser pulses, detecting the resulting acoustic signals and converting these signals into an image using reconstruction algorithms. A good reconstruction algorithm can allow accurate visualization of complex anatomical features, and also facilitate further multispectral analysis. This dissertation describes various model-based reconstruction algorithms for optoacoustic tomography.
Model-based reconstruction is generally more accurate than reconstruction based on analytical inversion, but it requires more computational and memory resources. Here, a much faster optoacoustic reconstruction method is proposed, in which the model matrix and the optoacoustic signal are transformed into the wavelet domain. Pseudoinverse of model matrices can be calculated on a much smaller scale, and then multiplied with the corresponding signals to form the final optoacoustic image. Using this methodology over an order of magnitude reduction in inversion time is demonstrated for numerically generated and experimental data.
Second, sparsity-based reconstruction is developed for a 2D optoacoustic imaging system. Specifically, a cost function is used that includes the L1 norm of the image in sparse representation along with a total variation term. The minimization process is implemented using gradient descent with backtracking line search. This algorithm leads to sharper reconstructed images with weaker streak artifacts than either conventional L2-normregularized reconstruction or back-projection reconstruction with simulated and experimental datasets.
Next, the sparsity-based reconstruction is adapted to 3D geometries, thereby exploiting more of the potential of tomography because the ultrasound waves generated after sample illumination propagate in all directions. To accelerate the reconstruction, Barzilai-Borwein line search is used to analytically determine the step size during gradient descent optimization. The proposed method offers 4-fold faster reconstruction than the previously reported L1-norm regularized reconstruction based on gradient descent with backtracking line search. The new algorithm also provides higher-quality images with fewer artifacts than L2-norm regularized reconstruction or back-projection reconstruction.
Finally, this dissertation develops frequency domain methods for reconstructing optoacoustic images when the sample is illuminated with an amplitude-modulated continuous-wave laser. The numerical method is used to guide the design of experimental set-ups for optoacoustic tomography in the frequency domain, as well as the selection of measurement parameters.
The methods developed in this dissertation enable robust processing and inversion during optoacoustic reconstructions, which may enhance the performance of optoacoustic imaging and tomography in preclinical and clinical environments, as well as open up avenues for further theoretical and experimental developments.
«
Optoacoustic tomography can generate high-resolution optical images of biological samples in vivo at depths of several millimeters to centimeters. The technique is based on illuminating the sample with nanosecond laser pulses, detecting the resulting acoustic signals and converting these signals into an image using reconstruction algorithms. A good reconstruction algorithm can allow accurate visualization of complex anatomical features, and also facilitate further multispectral analysis. This di...
»