Photoacoustic tomography reconstructs images from acoustic signals generated by light absorption, but the reconstruction typically requires knowledge of the sound speed in the tissue. When that speed varies spatially and is unknown, the problem becomes more challenging. The new paper on arXiv considers the joint reconstruction of the initial pressure distribution and the spatially varying sound speed from boundary measurements.
The authors propose a regularized data-driven framework to address this joint inverse problem. According to the abstract, the framework incorporates an untrained CNN and coefficient-space regularization, as indicated by the title. The approach is designed to handle the ill-posedness that arises when both fields are unknown.
The abstract does not include experimental results or implementation details, so the specific benefits and performance of the method are not yet available from this summary. The full paper presumably provides the mathematical formulation and validation.