Our group has concentrated on development of a 3D photoacoustic imaging system for biomedical imaging research.
The technology employs a sparse parallel detection scheme and specialized reconstruction software to obtain 3D optical
images using a single laser pulse. With the technology we have been able to capture 3D movies of translating point
targets and rotating line targets. The current limitation of our 3D photoacoustic imaging approach is its inability ability
to reconstruct complex objects in the field of view. This is primarily due to the relatively small number of projections
used to reconstruct objects. However, in many photoacoustic imaging situations, only a few objects may be present in
the field of view and these objects may have very high contrast compared to background. That is, the objects have
sparse properties. Therefore, our work had two objectives: (i) to utilize mathematical tools to evaluate 3D photoacoustic
imaging performance, and (ii) to test image reconstruction algorithms that prefer sparseness in the reconstructed images.
Our approach was to utilize singular value decomposition techniques to study the imaging operator of the system and
evaluate the complexity of objects that could potentially be reconstructed. We also compared the performance of two
image reconstruction algorithms (algebraic reconstruction and l1-norm techniques) at reconstructing objects of
increasing sparseness. We observed that for a 15-element detection scheme, the number of measureable singular vectors
representative of the imaging operator was consistent with the demonstrated ability to reconstruct point and line targets
in the field of view. We also observed that the l1-norm reconstruction technique, which is known to prefer sparseness in
reconstructed images, was superior to the algebraic reconstruction technique. Based on these findings, we concluded (i)
that singular value decomposition of the imaging operator provides valuable insight into the capabilities of a 3D
photoacoustic imaging system, and (ii) that reconstruction algorithms which favor sparseness can significantly improve
imaging performance. These methodologies should provide a means to optimize detector count and geometry for a
multitude of 3D photoacoustic imaging applications.
|