Self-Supervised Z-Slice Augmentation for 3D Bio-Imaging via Knowledge Distillation

Dataset used to train ZAugNet, a neural network for Z-slice augmentation, that encompasses a variety of shapes, textures, and microscopy techniques, as described below:

Ascidian Embryos: This dataset consists of 3D confocal images of P. mammillata embryos, captured using fluorescence microscopy. The plasma membrane was imaged using a PH::Tomato construct, and images were taken at 20°C with a Leica TCS SP8 inverted microscope, resulting in cubic voxel datasets. Credits: Rémi Dumollard, Alex McDougall.

Cell Nuclei: This dataset includes 3D confocal images of colorectal cancer organoids, stained with DAPI. The images were captured using a Nikon Spatial Array Confocal (NSPARC) detector with 40x objective, providing high-resolution data on organoid structures. Credits: Yekaterina A. Miroshnikova.

Filaments of Microtubules: This dataset features 3D images of microtubules in Mouse Embryonic Fibroblasts, captured using a Zeiss LSM 900 Airyscan2 with a high-resolution 63x oil objective. The images focus on the microtubule network and were post-processed using Airyscan technology. Credits: Benoit Vianay, Alexandra Colin.

Human Embryos: This dataset contains a time-lapse image of a human embryo obtained as part of in vitro fertilization procedure, captured using an EmbryoScope Plus incubator. The images were acquired across 11 focal planes at 15-minute intervals, providing detailed temporal data of embryo development. Credits: Elsa Labrune.

These datasets offer diverse microscopy techniques and biological subjects, supporting a variety of training scenarios for neural networks. Additional information about these datasets can be found in the Methods section of the associated paper.