Fluorescence microscopy is one of the most widely used tools in biology and medicine because it allows researchers to label specific molecules and visualize structures inside cells and tissues with high contrast. But there is a tradeoff. High numerical aperture objectives provide high spatial resolution, yet they also have a very shallow depth of field. In thick samples, only a thin layer stays in focus, while structures above and below quickly become blurred. In multicolor imaging, the challenge becomes even harder because different wavelengths can focus differently across depth. As a result, thick and multi-label specimens often require axial scanning, which increases acquisition time and can introduce motion artifacts, photobleaching, and phototoxicity.
In a new paper published in Light: Science & Applications , Professor Humeyra Caglayan and her colleagues from Tampere University, INRIA Rennes, and Eindhoven University of Technology developed MANTIS to address these limitations. The system combines a learned titanium dioxide meta-optic with a neural reconstruction network trained end to end. The meta-optic is placed at the Fourier plane of a 4f relay system, where it reshapes the optical response across both depth and wavelength. The captured image is then processed computationally to reconstruct a sharp multispectral result, without axial scanning, extra corrective optics, or separate calibration for each color channel.
The researchers summarize the core idea of the system as follows: “We realize this concept in MANTIS, a co designed computational microscopy system that addresses both extended depth of field and chromatic aberration through joint optimization of a learned meta-optic and a neural reconstruction model.” They explain that, in a conventional high numerical aperture microscope, defocus and chromatic dispersion make the point spread function change rapidly with axial position and wavelength. In MANTIS, the learned phase profile reshapes the pupil so that image information is preserved more robustly under defocus across multiple spectral channels, improving the conditioning of the reconstruction problem.
The team experimentally demonstrated a 50 micrometer extended depth of field at numerical aperture 1.1, corresponding to an 82-fold increase compared with a conventional wide field microscope. In simulations, the method also targeted depth ranges from 25 to 75 micrometers, showing the expected tradeoff between larger depth range and reconstruction fidelity. Measurements showed reduced axial sensitivity and improved chromatic stability, while reconstructed images preserved contrast and lateral detail across depth.
To test the method in realistic biological conditions, the researchers imaged MDCK epithelial monolayers, three dimensional MDCK II spheroids about 50 micrometers thick, and a 16 micrometer mouse kidney section. In spheroid imaging, MANTIS preserved cell boundaries and resolved nuclei more clearly throughout the volume than conventional wide field microscopy. In kidney tissue, it recovered fine tubular and cytoskeletal structures that were not visible in conventional images taken at the same out of focus planes. Together, these results show that MANTIS can deliver high resolution, cross channel consistent multispectral fluorescence imaging of thick biological specimens in a single shot, offering a practical route toward faster imaging without mechanical refocusing.
Light: Science & Applications
Multispectral Extended Depth-of-Field Fluorescence Microscopy with Co-designed Meta-Optics and Neural Reconstruction