Photoacoustic tomography represents a hybrid modality that synergistically integrates optical excitation and ultrasonic detection to visualize optical absorption within biological tissues. By being compatible with clinical linear array ultrasound transducers, photoacoustic imaging demonstrates substantial potential for clinical implementation. However, the restricted field of view inherent to linear transducers may compromise image quality and give rise to artifacts.

To address these limitations, we initially developed a multi-view imaging system designed to restore images under limited-view conditions. To expedite and overcome the practical challenges associated with reconstructing full-view images, we introduced a deep learning framework predicated on a transformer network incorporati...[ Read more ]

Photoacoustic tomography represents a hybrid modality that synergistically integrates optical excitation and ultrasonic detection to visualize optical absorption within biological tissues. By being compatible with clinical linear array ultrasound transducers, photoacoustic imaging demonstrates substantial potential for clinical implementation. However, the restricted field of view inherent to linear transducers may compromise image quality and give rise to artifacts.

To address these limitations, we initially developed a multi-view imaging system designed to restore images under limited-view conditions. To expedite and overcome the practical challenges associated with reconstructing full-view images, we introduced a deep learning framework predicated on a transformer network incorporating neighborhood attention mechanisms. This novel method captures both local and long-range pixel dependencies, enabling the reconstruction of high-resolution images from sparse, limited-angle input data. The promising results underscore the potential of our deep learning strategy to surmount the limited-data challenge and facilitate high-fidelity photoacoustic imaging with linear arrays. Preliminary evaluations on hybrid data sets reveal that our approach achieves state-of-the-art performance in limited-view reconstruction when compared to others conventional restoration task models