SAS-Net: Scene-Appearance Separation Network for Robust Spatiotemporal Registration in Bidirectional Photoacoustic Microscopy

Researchers introduced SAS-Net, a novel AI architecture that separates scene content from appearance characteristics to correct severe image misalignment in high-speed, bidirectional photoacoustic microscopy (OR-PAM). The model achieved a normalized cross-correlation score of 0.961 and processes frames in 11.2 ms (86 FPS), enabling real-time analysis. An ablation study showed removing a key component caused an 82% performance drop, highlighting its critical design. The code will be made publicly available.