We are very excited to release the new website (link) with all the information about CEREBRUM 7T, an optimised end-to-end Convolutional Neural Network (CNN) architecture, that allows for the segmentation of a whole 7T T1w MRI brain volume at once. Training is performed in a weakly supervised fashion, exploiting labelling with errors obtained with automatic state-of-the-art methods. The trained model is able to produce accurate multi-structure segmentation masks on six different classes in only a few seconds.
Above, reconstructed meshes of GM, WM, basal ganglia, ventricles, brain stem, and cerebellum of a testing volume, obtained with CEREBRUM7T on sub-013_ses-001. A light smoothing operation is performed (50 iterations – BrainVoyager) – no manual corrections.
In the website, we release the code, we make available the entire dataset used in the paper, we add instructions on how to use the containers in docker and singularity, and we display a lot of results.
Visit https://rocknroll87q.github.io/cerebrum7t/ for more.