Learning-based sound-speed correction for dual-modal photoacoustic/ultrasound imaging

Accurate determination of the speed-of-sound (SoS) within the propagation medium is of significant importance in photoacoustic (PA) image reconstruction. A common practice in PA imaging assumes a homogeneous SoS distribution, e.g., 1540 m/s for soft tissue similar to that implemented in conventional ultrasound (US) beamforming. This assumption can lead to US aberration artefacts that degrade the image quality due to tissue heterogeneity. In this work, we introduce a learning-based method focusing on compensating SoS variations for PA image reconstruction in a dual-modal PA/US imaging system. Deep neural networks were trained for SoS retrieval using US channel data and subsequently informed the corresponding PA image reconstruction. The proposed framework demonstrated effective mitigation of US aberration artefacts with a numerical phantom, achieving a structural similarity index measure of 0.8267 compared to 0.5042 with the conventional SoS assumption of 1540 m/s. Likewise, the enhancements were also evident when testing the framework with ex vivo US/PA data, implying its great potentials in improving PA image quality for in vivo applications.