MorphoGraphX has a CNN AddOn that can be used to process stacks and recognize structures by using convolutional neural network deep learning tools. This can greatly improve downstream segmentation results. The AddOn requires Cuda and is available on our Software downloads page.
Currently the CNN AddOn ships with 3 model files, all of which are based on the 3D UNet architecture:
WillisUNet.pt – This file was trained with the Cell Segmentation Software from the Stegmaier lab based on the Willis et al. 2016 dataset from the Jönsson lab available online here: https://www.repository.cam.ac.uk/handle/1810/262530.
VijayanUNet.pt – This file was trained with the pytorch-3dunet software based on data from Vijayan et al. 2021 from the Schneitz lab. MorphoGraphX was used to segment and curate the training data for this network.
BasselCombinedUNet.pt – This file was trained with the pytorch-3dunet software based on the the Willis and Vijayan data sets above combined with data from the George Bassel lab.
In addition to running pre-trained networks, MorphoGraphX has tools to curate and correct 3D datasets, including operations such as merging and splitting cells, and other editing tools for 3D voxel data.
MorphoGraphX also has tools for cell classification on 2.5D and 3D meshes using Support Vector Machines, and can incorporate positional information as cell attributes.
To see Deep MGX in action see our YouTube videos
MorphoGraphX 