Photoacoustic computed tomography (PACT) is a promising technique for functional brain imaging because it provides label-free hemodynamic imaging with deep penetration, high imaging speed, and superior image quality. A three-dimensional PACT (3D-PACT) system has been developed, capable of producing high-resolution, multipurpose images for applications ranging from rodent brain studies to human breast imaging. However, reconstructing high-quality 3D images in PACT is time-consuming, requiring subjects to remain still, which limits practical applications such as breast imaging, where breath-holding for extended periods is challenging. We propose accelerating 3D-PACT using a deep learning-based neural operator called PACTNO that significantly reduces the required scanning time. PACTNO’s architecture is agnostic to the number of sensors used in the system, making it versatile for various imaging setups. We demonstrate the robustness and efficacy of PACTNO in reconstructing high-quality images from real phantom data, achieving consistent performance even when the number of transducers is reduced or with limited-angle acquisition. This advancement holds promise for making 3D-PACT more accessible and practical for clinical use.
@article{wang2024ultrasound,
title={Neural Operators Accelerate 3D Photoacoustic Computed Tomography},
author={Wang, Jiayun and Aborahama, Yousuf and Berner, Julius and Li, Zongyi and Zhang, Yang and Azizzadenesheli, Kamyar and Wang, Lihong V. and Anandkumar, Anima},
year={2024}
}