Compression artifact removal in photoacoustic tomography based on transformer with enhanced self-attention

Photoacoustic tomography is an emerging biomedical imaging technique that has demonstrated broad application prospects in various biomedical fields. With the increasing demand for high-resolution and real-time imaging, the volume of data has grown rapidly, presenting significant challenges for data storage and processing. While compression methods alleviate the burden, it often introduces noticeable artifacts, making it hard to balance data reduction and high-quality imaging. Therefore, a novel approach is required to mitigate image distortion and artifacts introduced by data compression. This article presents a compression artifact removal in photoacoustic tomography based on Transformer with enhanced self-attention. The proposed framework confines self-attention computation to local windows and enables cross window information exchange through a shifted window translation, significantly enhancing computational efficiency while preserving strong global modeling capacity. Experiments on both simulated and in vivo mouse data demonstrate that the method effectively suppresses photoacoustic tomography compression artifacts and restores fine structural details. Quantitative evaluations indicate that the photoacoustic images reach a peak signal-to-noise ratio of 34.37 and a structural similarity index of 0.9178 at compression ratios up to 1:40, significantly mitigating the detail loss and artifacts introduced by data compression. In summary, the proposed method significantly enhances the image quality and accuracy of photoacoustic tomography, demonstrating great potential for practical applications in biomedical imaging.