To achieve high-resolution photoacoustic tomography in 3D requires either a high-density 2D detection array or multiple sequential acquisitions (scanning). Due to the cost and fabrication challenges of suitable 2D arrays, scanning systems are often used. The drawback is that scanning leads to long acquisition times and a trade-off must be made between image resolution and frame rate. This paper describes how techniques from compressed sensing can ameliorate this problem.
The main idea of compressed sensing is that a fully sampled data set usually contains redundant information, and a high-quality image can be recovered from much less data by exploiting the low spatial, temporal and/or spectral complexity of the chromophore distributions. Scanners that are able to suitably sub-sample the acoustic field can acquire sufficient data much faster.
To reconstruct high-quality images from sub-sampled data, iterative, model-based approaches incorporating explicit constraints on the characteristics of the chromophore distributions are used. Such algorithms can be computationally demanding, but also highly versatile. They are applicable to all scanning geometries, can be extended to incorporate complex acoustic models accounting for heterogeneous media, and could even include optical models for quantitative reconstructions.
Experimental demonstrations of both static in-vivo and dynamic imaging will be described, using data obtained with a sequential scanner. The static images can be obtained 4x-8x times faster using this approach. A further increase 2x increase can be achieved when imaging dynamic processes and using motion estimation models. Finally, dynamic, high-resolution 3D PAT imaging with frame rates exceeding 1Hz will be demonstrated experimentally.
|