Real-time robust autofocus method enabling sustained intravital scanning light field imaging.

Recent advances in computational microscopy enable highspeed high-resolution intravital 3D imaging with low phototoxicity. However, inevitable sample vibration and tissue deformation in multi-cellular organisms make it extremely challenging to maintain samples stably in focus over long-term even with an extended effective depth of field. Here, we propose a real-time robust autofocus method based on scanning light-field microscopy (AFsLF), enabling sustained high-speed 3D imaging of diverse samples across several days by continuously tracking the sample focal plane without hardware modifications. Based on the intrinsic disparity of light-field angular measurements, AFsLF estimates the focal plane with less than 2 µm error over a 500 µm depth range, completing within 0.1 s, 300-time faster than previous methods. We validate AFsLF across diverse tissues and challenging conditions, including low excitation power, multichannel illumination, and large axial displacements, enabling stable, long-term, multichannel subcellular imaging of neural activities and immune responses in mouse brain and liver.