GAP-Net-Based Real-Time Phase Reconstruction for Dynamically Evolving Holographic Scenes in Bioprinting

Holographic phase enables quantitative observation of the morphology of bioprinted structures and holds strong potential in bio-fabrication. However, achieving accurate and rapid holographic phase reconstruction for real-time monitoring during 3D bioprinting remains a challenge. In this paper, we propose a novel Global Attention Phase Network (GAP-Net) that focuses on real-time and accurate phase reconstruction for holographic phase monitoring in bioprinting. The GAP-Net leverages lightweight feature extraction module and global attention for semantic segmentation of wrapped phases, enhancing reconstruction precision. In addition, by predicting the first six Zernike coefficients to capture the dominant background aberrations, the method effectively mitigates phase distortions and further improves reconstruction accuracy. The approach achieves real-time phase reconstruction at 22.8 FPS with a 1.11-fold accuracy improvement over existing deep learning methods. The proposed method also enables successful real-time visualization of the curing dynamics of biological samples during bioprinting, further demonstrating its practical effectiveness. This work not only provides real-time observational data for bioprinting but also shows promising potential for broader applications in fields such as optics and biomedical imaging.