Residual-conditioned sparse transformer for photoacoustic image artifact reduction

Abstract

Photoacoustic tomography (PAT) combines the high spatial resolution of ultrasound imaging with the high contrast of optical imaging. To reduce acquisition time and lower the cost of photoacoustic imaging, sparse sampling strategy is often employed. Conventional reconstruction methods often produce artifacts when dealing with sparse data, affecting image quality and diagnostic accuracy. This paper proposes a Residual-Conditioned Sparse Transformer (RCST) network for reducing artifacts in photoacoustic images, aiming to enhance image quality under sparse sampling. By introducing residual prior information, our algorithm encodes and embeds it into local enhancement and detail recovery stages. We utilize sparse transformer blocks to identify and reduce artifacts while preserving key structures and details of the images. Experiments on multiple simulated and experimental datasets demonstrate that our method significantly suppresses artifacts and improves image quality, offering new possibilities for the application of photoacoustic imaging in biomedical research and clinical diagnostics.