PyTorch implementation of DeepTaskGen, a convolutional neural network designed to predict whole-brain voxel-wise individual task contrast maps from voxel-to-ROI resting-state connectomes. The architecture adapts and further extends BrainSurfCNN (Ngo et al., 2022) for volumetric representation of the brain, incorporating features such as attentional gates and Contrast-Regularized Reconstructive Loss.
Scripts for training DeepTaskGen and validating it on three separate datasets (HCP-YA, HCP-D, and UKB) are available at https://github.com/eminSerin/deeptaskgen-paper.
Pretrained models can be found at https://github.com/eminSerin/deeptaskgen-models.
- Ngo, Gia H., et al. "Predicting individual task contrasts from resting‐state functional connectivity using a surface‐based convolutional network." NeuroImage 248 (2022): 118849.
- Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. "U-net: Convolutional networks for biomedical image segmentation." International Conference on Medical image computing and computer-assisted intervention. Springer, Cham, 2015.
- Oktay, Ozan, et al. "Attention u-net: Learning where to look for the pancreas." arXiv preprint arXiv:1804.03999 (2018).
Funded by the European Union (Grant agreement No 101057429). Complementary funding was received by UK Research and Innovation (UKRI) under the UK government’s Horizon Europe funding guarantee (10131373 and 10038599) and the National Key R&D Program of Ministry of Science and Technology of China (MOST 2023YFE0199700). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union, the European Health and Digital Executive Agency (HADEA), UKRI or MOST. Neither the European Union nor HADEA nor UKRI nor MOST can be held responsible for them.
