Multi-bandwidth reconstruction for photoacoustic tomography using cascade U-net

Photoacoustic imaging (PAI) employs short laser pulses to excite absorbing materials, producing ultrasonic waves spanning a broad spectrum of frequencies. These ultrasonic waves are captured surrounding the sample and utilized to reconstruct the initial pressure distribution tomographically. Despite the wide spectral range of the laser-generated photoacoustic signal, an individual transducer can only capture a limited segment of the signal due to its constrained bandwidth. Herein, we have developed a multi-bandwidth ring array photoacoustic computed tomography (PACT) system, incorporating a probe with two semi-ring arrays: one for high frequency and the other for low frequency. Utilizing the two semi-ring array PAIs, we have devised a specialized deep learning model, comprising two serially connected U-net architectures, to autonomously generate multi-bandwidth full-view PAIs. Preliminary results from simulations and in vivo experiments illustrate the system's robust multi-bandwidth imaging capabilities, achieving an excellent PSNR of 34.78 dB and a structural similarity index measure (SSIM) of 0.94 in the high-frequency reconstruction of complex mouse abdominal structures. This innovative PACT system is notable for its capability to seamlessly acquire multi-bandwidth full-view PAIs, thereby advancing the application of PAI technology in the biomedical domain.