Temporal variance mapping with machine learning for label-free 3D chromatin imaging using optical interferometric microscopy.

Label-free cell imaging using phase-sensitive optical interferometric microscopy enables noninvasive observation of living cells, but it often suffers from low imaging specificity and limited spatial resolution, particularly in the axial direction. In this study, we present a label-free method for high-resolution 3D chromatin imaging by leveraging rapidly fluctuating scattering signals arising from native biomolecular motions, captured using a high-speed and highly sensitive interferometric microscope. Optical transmission images of live cell nuclei are recorded at 1000 frames per second, and temporal variance maps are computed from these recordings. Deep learning models are then trained to map the label-free dynamics data to confocal fluorescence images of chromatin. Our results demonstrate that the resulting dynamics maps resolve fine subnuclear structures, including nucleoli and nuclear speckles-the latter being especially difficult to detect using conventional phase microscopy. Notably, the use of second-order temporal statistics leads to significantly enhanced axial resolution, enabling effective 3D imaging of chromatin architecture. This work highlights the potential of temporal signal analysis in fast, label-free optical interferometric microscopy and paves the way for broader applications in high-resolution, label-free imaging of dynamic biological structures.