A multimodal optical microscopy dataset for characterizing cytoskeletal organization.

Optical microscopy combined with machine learning for image analysis is seeing increasing use for studying cell morphology and the organization of subcellular structures such as the cytoskeleton. Training and validating machine learning models require high-quality training data, but datasets that combine multiple modes of imaging for studying cytoskeletal structures are lacking. Here, we present the first curated multimodal imaging dataset consisting of 5 imaging modalities and 11 imaging channels, designed to support quantitative analysis of cytoskeletal organization and cell morphology. The dataset comprises 2253 individual HeLa cells imaged with brightfield, Reflection Interference Contrast Microscopy (RICM), widefield fluorescence, Total Internal Reflection Fluorescence microscopy (TIRFm), and confocal microscopy. Cells are stained for key cytoskeletal components (actin filaments, focal adhesions, and microtubules) across 4 cytoskeletal treatment conditions, enabling the comparison of changes to cytoskeletal organization at the subcellular level. The dataset includes both single-plane multichannel images, and multi-plane z-stacks, supporting two-dimensional analysis as well as three-dimensional reconstruction. The dataset is accompanied by metadata describing imaging conditions, channel alignment, and acquisition parameters. Additionally, annotated masks are provided for cell nuclei and footprint as ground truths. Validation is performed using quantitative cell shape descriptors and benchmarking of automated segmentation approaches for cytoskeletal structures. Overall, multiple imaging modalities with consistent labeling make this dataset valuable for studying cytoskeletal organization, cell morphology, and mechanobiology.