DeepProjection: Specific and robust projection of curved 2D tissue sheets from 3D microscopy using deep learning
Code repository for publication: Haertter, D. et al., Development (2022) doi:10.1242/dev.200621.
Comparison of DeepProjection with maximum intensity projection. A: Maximum intensity projection (MIP) of a single stack (8 slices, 1 µm z-distance) of images of the dorsal opening of a Drosophila embryo during dorsal closure, cell boundaries labeled with Cadherin-GFP. B: MIP of as single stack (53 slices, 2 µm z-distance) of images of a zebrafish periderm labeled with labeled krt4-lyn-GFP. A’-D’: y-z cuts of 3D image stacks at red dashed line in A-B. C’ and D’ show the masked stack with the manifolds predicted by DP. C, D: DeepProjection (DP) results from the same stacks of Drosophila and zebrafish embryo. E, F: Zoom into amnioserosal tissue in Drosophila comparing MIP (E) and DP (F), showing successful masking of yolk granules and gut tissue underneath the amnioserosal tissue. G, H: Zoom into zebrafish embryo comparing MIP (G) and DP (H), showing masking of underlying epithelial tissue layer. Scale bars: A, C 50 µm; B, D 100 µm; E-F 10 µm; G, H: 50 µm.
IMPORTANT: Convolutional neural networks run much faster on NVIDIA-GPUs than on CPUs. To enable training and prediction on GPUs, users need to install CUDA toolkit and the corresponding version of PyTorch from the official PyTorch website. Select the correct version of CUDA on this webpage and run the command in your terminal. DeepProjection was tested with CUDA 11.0 and Pytorch 1.7.1.
pip install deepprojection for current version. pip install deepprojection==0.0.10for newest version with old API.
- Download the code using the GitHub .zip download option or clone repository.
- Install all Python packages listed in requirements.txt
import deepprojection as dp
A detailed instruction on how to train and use DeepProjection can be found here. Training data, test data and pretrained networks for Drosophila dorsal closure and zebrafish periderm used in publication can be found here.
Run GUI in terminal by python your_path/deepprojection/GUI.py. Make sure to use an environment with all necessary packages installed.
For question or suggestions, please create an issue on GitHub or send us a mail.
This work is licensed under the GPL3 License.