3D photoacoustic computed tomography enhanced by 3D progressive U-shaped enhancement network (3D-pU-net)

Abstract

Photoacoustic computed tomography (PACT) has emerged as a practical tool for fast 3D imaging with optical contrast that give morphological, functional, and molecular in vivo information. The spatiotemporal resolution of the PACT system are decided by the composed hardware specification. Hence, to achieve better image quality and faster imaging speed, the high-specification hardware should be supported, but it leads to huge costs. Here, we propose a new solution to overcome the inherently trade-offs between imaging speed and image quality based on a neural network, a 3D progressive U-shaped enhancement network (3D-pU-net). In our approach, a hemispherical transducer array-based PACT system was used for the system configuration, and we could obtain accurate high-quality reference images with all elements of the array. Cluster sampling, which was used for input data, is not affected by imaging speed degradation, but the image quality is degraded. We demonstrated that the trained 3D-pU-net enhanced the image quality of cluster-sampled data of static whole-body imaging. Furthermore, the network also performed a wide range of applications such as dynamic observation of contrast agent kinetics. In this study, we showed that the 3D-pU-net could improve the anatomical contrast and spatial resolution by overcoming the limited-view effect. This proposed approach can help a variety of PACT applications in practical settings, allowing for the development of useful and economical imaging equipment.