Fused deep learning enables 6D single-molecule localization in polarization-resolved microscopy.

Single-molecule orientation localization microscopy (SMOLM) is an optical means to measure complex transport in charged and crowded conditions, such as inside cells or polymer materials. SMOLM extracts time- and space-dependent three-dimensional orientation information from dipole emitters. Achieving simultaneous position-orientation resolution with high photon efficiency remains a central challenge in SMOLM instrument design. We developed an optical fluorescence microscope that uses the double-helix point spread function (DHPSF) to localize dipole emitters in six dimensions (6D), delineated by spatial and dynamic orientational parameters. Furthermore, we developed a fused deep learning approach based on existing neural network architectures to localize dipole emitters in 6D. Our microscope enables simultaneous 6D localization of single fluorophores, achieving a median spatial precision of 10 nm and angular precision below 10° across most of orientation space, except for the azimuthal angle at high polar angles where the DHPSF exhibits known optical degeneracies. We demonstrate our approach by localizing single rhodamine B molecules in poly(methyl methacrylate) films. The recovered orientations shownear 90° and small wobble angles. We also demonstrate 6D SMOLM of a spherical supported lipid bilayer, where despite the low signal, out-of-training distribution of the experimental data, we observe clearly ordered orientation of Nile red molecules within the membrane.