This repository provides a tool for generating task-based fMRI functional connectivity matrices from resting-state fMRI (rs-fMRI) using image-to-image mapping. The primary objective is to create task-related connectivity matrices, simulating different tasks, as included in the Human Connectome Project (HCP) dataset.
Figure 1: Overview of the connectivity matrices estimation process, showing the workflow from raw data input through preprocessing steps.
While task-based fMRI is essential for understanding brain function during specific activities, collecting task-based data can be resource-intensive and time-consuming. This tool leverages resting state fMRI data and mapping techniques from Optimal Transport theory to create task-related fMRI data, offering a cost-effective and efficient approach for researchers.
Figure 2: Schema illustrating the transfer of fMRI patterns to task-specific domains.
- Task Generation: Transform resting-state fMRI connectivity matrices into task-specific fMRI data for multiple tasks (e.g., EMOTION, GAMBLING, LANGUAGE, MOTOR, RELATIONAL, Working Memory, SOCIAL) from the HCP dataset.
- Customization: Adjust parameters to fine-tune the task generation process.
To get started with using this tool, follow the instructions provided in the readme.md and ensure you have the required dependencies installed.
- Data Preparation: Ensure you have the HCP resting-state fMRI connectivity matrices as input data.
- Configuration: Adjust the tool's parameters as needed, specifying the desired task (e.g., EMOTION, GAMBLING) and other options.
- Run the Generator: Execute the notebooks to generate task-based fMRI data.
| Model Score | VAE | cGAN | WcGAN-QC | Vanilla NOT |
|---|---|---|---|---|
| MSE score | 0.07 ± 0.25 | 0.06 ± 0.01 | 0.05 ± 0.01 | 0.06 ± 0.03 |
| L1 | 0.69 ± 0.25 | 0.19 ± 0.01 | 0.18 ± 0.01 | 0.22 ± 0.03 |
Table 1: Comparison of reconstruction quality metrics across different generative models
This project is licensed under the Skoltech Academic License.
For questions, suggestions, or issues, please feel free to contact us.