|
Photoacoustic Tomography (PAT) is a useful tool for fast 3D imaging that provides structural, molecular, and functional in vivo information. It is capable of producing 3D images using a multi-element hemispherical array transducer. PAT images can be enhanced a great number of ultrasonic transducer components with multiplexers, but this can result in high costs and slow temporal resolution because of using multiplexers. In this research, we present a deep learning solution to improve both the spatial and temporal resolution in PAT. We demonstrated that the trained neural network enhanced the image quality of a quarter-cluster-sampled data of static whole-body imaging. Our approach increased limited-view aperture and the spatial resolution by around three and two times, respectively. Additionally, it allowed to improve temporal resolution by four times without multiplexing. Our method also demonstrated excellent performance in contrast-enhanced PA imaging, enabling molecular imaging. Our strategy has the potential to enable high spatial and temporal resolution observation of biodynamics in 3D PAT without being limited by hardware constraints.
|
