Deep learning-based speed of sound aberration correction in photoacoustic images

Beamforming algorithms are widely used for photoacoustic (PA) imaging to reconstruct the initial pressure map. In the reconstruction process, they typically assumed that the imaged biological tissue was a homogeneous medium. However, as biological tissue is generally heterogeneous, the misassumption causes suboptimal image reconstruction. Because it is difficult to predict the heterogeneity of a medium, it was still common to reconstruct images assuming a uniform medium. To solve this problem, we introduce a deep learning-based algorithm that can correct the speed of sound (SoS) aberration in the PA image. We trained a neural network with the multiple simulation datasets and successfully corrected SoS aberrations in a PA in vivo image of the human forearm. We observed that the proposed algorithm effectively suppressed side lobes and noise in the PA image and greatly improves image quality.1