BioSR for LLS-SIM

BioSR for LLS-SIM is a biological image dataset acquired using our home-built lattice light-sheet structured illumination microscopy (LLS-SIM). It currently includes paired diffraction-limited LLSM and LLS-SIM images of a variety of biology structures, constituting a high-quality volumetric SR dataset. The BioSR for LLS-SIM dataset originates from our paper [Qiao, C., Li, Z., Wang, Z., Lin, Y., ... & Li, D. (2024). Fast adaptive super-resolution lattice light-sheet microscopy for rapid, long-term, near-isotropic subcellular imaging. bioRxiv, 2024-05], serving as the supplementary data to execute meta-training or other proposed deep-learning-based algorithms, which provides sources for the community to try our methods, replicate our results, and develop their own methods for super-resolution lattice light-sheet microscopy.

The BioSR for LLS-SIM dataset includes 10 diverse task datasets from 10 distinct biological specimens (granular component, chromosomes, fibrillar center, fibrillarin, Lyso, MTs, F-actin in pollen tubes, inner mitochondrial membrane, ER in adherent Cos-7 cells and ER in mitotic Hela cells during metaphase). For each type of samples, we acquired raw LLS-SIM images from about 30-50 ROIs. For each ROI, five different levels of light intensity ranging from low to high fluorescence levels were acquired, and the images of the highest fluorescence level (i.e. the GT raw images) were reconstructed into high-quality GT LLS-SIM images via the conventional LLS-SIM reconstruction algorithm, which could be used as the groud truth in the training phase of deep-learning models. See training dataset.xls file for more specific information of the training datasets.

In this dataset, we also provide representative testing data, as well as the pre-trained meta-model and several representative finetuned models. See testing dataset and pre-trained models.xls file for more specific information of the demo testing data and pre-trained models.

The testing dataset and models were uploaded on the current repository, while we assigned each training dataset of a specific biological specimen to an exclusive repository with its own DOI and URL due to the quota policy of Zenodo, which are listed as follows:

Structure
URL

1
Granular Component
https://doi.org/10.5281/zenodo.14348076

2
Chromosomes
https://doi.org/10.5281/zenodo.14351353

3

Innermost Fibrillar Center

https://doi.org/10.5281/zenodo.14351668

4
Fibrillarin
https://doi.org/10.5281/zenodo.14373971

5
Lysosome
https://doi.org/10.5281/zenodo.14351377

6
Microtubules

https://doi.org/10.5281/zenodo.14373993

7
F-actin in pollen tubes
https://doi.org/10.5281/zenodo.14374023

8
Inner Mitochondrial Membrane
https://doi.org/10.5281/zenodo.14374065

9
ER in adherent Cos-7 cells
https://doi.org/10.5281/zenodo.14373955

10
ER in mitotic Hela cells during metaphase
https://doi.org/10.5281/zenodo.14436099