NeuroSeq

Proposal Abstract

There are many data sets with functional MRI which have some corresponding associated time-series data. How can functional MRI be used to predict this corresponding time series? For example, using the Forrest Gump data set, can we create a Seq2Seq style model which can predict dialogue from the movie, eye movements, or which can provide reliable object detection on stills from the film? We could use ICA timecourses, FNC state sequences, or raw fMRI as input to one end of the model. Can we do the inverse as well? E.g. can we predict neural information such as FNC states, ICA timecourses, or even full fMRI volumes from the movie data?

About the Project

Umbrella Project

• NeuroNeural

Emphasis:

• Neuroimaging/Data Science

Expected Background

• Graduate student or motivated undergraduate with ample time. Programming experience with python, especially for deep learning applications. Experience with deep neural networks in application.

Primary Point of Contact

• Brad Baker ([email protected])

Supervisor

• Sergey Plis ([email protected])

References and External Resources

• Catalyst: https://github.com/catalyst-team/catalyst
• Catalyst Neuro: https://github.com/catalyst-team/neuro
• Forrest Gump Data set: https://openneuro.org/datasets/ds000113/versions/1.3.0

Estimated Timelines

• Semester project

Possible Deliverables

• Set of Experiments with plots demonstrating the effectiveness of the model on HCP data set.
• 2-4 page report summarizing primary methodology and results.
• Eventual Submission to Conference or Workshop
• Eventual journal paper to Frontiers, NeuroImage