Spheroid LSFM dataset with nuclei centroid annotations

This dataset includes lightsheet images of spheroids of two different cell lines (LN18-RED and SH-SY5Y) and centroid annotations for model training and evaluation.If you use this dataset in your research, please cite our paper:

Van De Looverbosch, Tim, Sarah De Beuckeleer, Frederik De Smet, Jan Sijbers, and Winnok H. De Vos. “Proximity Adjusted Centroid Mapping for Accurate Detection of Nuclei in Dense 3D Cell Systems.” Computers in Biology and Medicine 185 (February 1, 2025): 109561. https://doi.org/10.1016/j.compbiomed.2024.109561.

The data is organized in the following way:LN18-RED
├── pretrain # Pretraining data with weak annotations
│ ├── patches
│ │ ├── sample_01
│ │ │ ├── patch_0000.tif
│ │ │ ├── ...
│ │ ├── ...
│ ├── patches_binary # Used to filter out patches with a lot of background
│ │ ├── sample_01
│ │ │ ├── patch_0000.tif
│ │ │ ├── ...
│ │ ├── ...
│ ├── points_csv # Weak centroid annotations from seeded watershed
│ │ ├── sample_01
│ │ │ ├── patch_0000.csv
│ │ │ ├── ...
│ │ ├── ...
SH-SY5Y
├── train # Training data with manual annotations
│ ├── patches
│ │ ├── sample_01
│ │ │ ├── patch_xxxx.tif
│ │ │ ├── ...
│ │ ├── ...
│ ├── points_csv
│ │ ├── sample_01
│ │ │ ├── patch_xxxx.csv
│ │ │ ├── ...
│ │ ├── ...
── test # Test data with manual annotations
├── patches
│ ├── sample_01
│ │ ├── patch_yyyy.tif
│ │ ├── ...
│ ├── ...
├── points_csv
│ ├── sample_01
│ │ ├── patch_yyyy.csv
│ │ ├── ...
│ ├── ...
preprocessed_images # Preprocessed in-toto images added for reference
├── LN18-RED
│ ├── sample_01.tif
│ ├── ...
├── SH-SY5Y
│ ├── sample_01.tif
│ ├── ...

1. Generation of spheroids and organoids

SH-SY5Y spheroid production was performed in U-bottom 96 well plates (Sigma-Aldrich, CLS7007-24EA) coated with anti-adherence solution for 20 minutes (Stemcell technologies, 07010). Spheroids from SH-SY5Y cells were left to form for at least three days before fixation. After seeding, the plates were centrifuged for three minutes at 100g before incubation at 37°C and 5 % CO2. Sample fixation was performed by adding 1:1 volume 4 % PFA (ROTI-Histofix 4 % paraformaldehyde, Roth, 3105.2) overnight. Samples were washed three times with PBS after fixation and stored in PBS + 0.1 % NaN3.

2. Light sheet fluorescence microscopy image acquisition

Prior to imaging, all samples (spheroids or organoids) were fixed in 4% paraformaldehyde, washed three times with PBS and stained with SiR-DNA (1.0 µM) nuclear counterstain (Spirochrome AG, SC007). Thereafter, the samples were embedded in 2% ultrapure agarose (Invitrogen, 16500500) using a custom-made cuboid mold. The resulting agarose cubes were placed in a glass-bottom dish (IBL Baustoff + Labor GmbH 220.110.021) and cleared overnight using the Fast 3D Clear protocol (Kosmidis et al., 2021) . The next day, the samples were mounted between the 5 mm water immersion TwinFlect mirrors (Leica, 158007011) of a Leica TCS SP8 Digital LightSheet microscope using a combination of a 2.5x illumination (506 523, fluostar, NA 0.07) and 10x detection objective (506 524 APO NA 0.30). We used a white light laser set at 628 nm for excitation of SiR-DNA. Images were acquired using a voxel size (X, Y, Z) of 0.3594 ⨉ 0.3594 ⨉ 1.9999 µm3.

3. Data preprocessing

For all datasets, the intensities of the nuclear (SiR-DNA) channel were normalized and clipped to the range [0.0, 1.0] using the 0.1 and 99.9 percentiles. No normalization was performed on available RFP and EGFP marker channels. To reduce image size, images were cropped based on the nuclear channel using the bounding box around each sample. For our LSFM data, the images were subdivided in slightly overlapping patches of 256 ⨉ 256 ⨉ 46 voxels, or approximately 92 ⨉ 92 ⨉ 92 µm3 using a stride of 224, 224 and 40 voxels along the X-, Y-, and Z-axis.

4. Ground truth annotation

For model training and evaluation, two sets of respectively 30 and 18 SH-SY5Y image patches were manually annotated. Manual annotation was performed in Napari by loading in a 3D image and creating a “points layer” to which annotated points, approximating the geometric center of the nucleus in X, Y and Z, were added.