High-fidelity super-resolution microscopy datasets spanning multispectral to hyperspectral domains via diffractive...

Paired datasets are critical for advancing data-driven microscopy but remain scarce for spectral imaging. Here, we present a comprehensive super-resolution dataset bridging multispectral and hyperspectral domains, generated via diffractive optics-based structured illumination. Unlike synthetic datasets derived from degradation models, these high-fidelity images are derived from physics-based modulation, ensuring accurate representation of biological structures. The collection is organized into five distinct records: (1) a 4-channel multispectral calibration dataset resolving ~96 nm details; (2) a 3-channel multispectral biological dataset of labeled organelles; (3) two spatial super-resolution datasets of cellular filaments reconstructed via pattern-illuminated Fourier ptychography and analytical phase-shifting to minimize artifacts; and (4) a novel hyperspectral dataset (503-689 nm, 6 nm interval, comprising 63 distinct groups) of bovine pulmonary artery endothelial cells acquired using a compact lattice SIM system. We provide paired diffraction-limited widefield (WF) and reconstructed super-resolution (SR) images, alongside system point spread functions. This dataset serves as a rigorous benchmark for developing image restoration, spectral unmixing, and cross-modality deep learning algorithms, facilitating the extraction of nanoscale insights from standard optical setups.