BME595 Dee[?]earning Project

Ninety days of drinking coffee(blue) or no drinking coffee(orange)

-blue is for drinking coffee days, orange is for not drinking coffee days. The different length of blue/orange color bars were represented different consecutive drinking coffee/not drinking coffee days.

Title

• Using "Neural Network" Prediction of bodily metabolism with MyConnectome dataset.

Team members

• NameB (GitHubUserB), Ho-Ching Yang (photoacoustic),

Goals

• Our project primarily focuses on applying convolutional neural network (CNN) to the problem of cross-scan (one subject, 90 scans) bodily metabolism prediction derived from human brain fMRI data.

Challenges

• Myconnectome dataset had recorded brain function and metabolism fluctuate in a single individual, Dr. Poldrack over the course of an entire year.
• There is 90 scans in this dataset. Each scans has 294 timepoint which means we'll have 90*294 3D fMRI image as input of Neural Network.
• The input will be 4D fMRI images (don't know if we need to count time into another dimension), in terms of the output will be bodily metabolism(ex. this is the functional MRI image scanned after this subject drinking coffee, eating breakfast...)
• "What" should we let neural network to "See" will be a big chanllenges. We should use processed dataset, which is convolved with hemodynamic response.
• In order to compare the package efficiency, we will learning new packages from

Restrictions

• I will be very happy if I knew a simplest way to train 4D fMRI dataset without GPU...

Reference

1. 
2. Nathawani, D., Sharma, T. and Yang, Y., 2016. Neuroscience meets deep learning.
3. Bastien, F., Lamblin, P., Pascanu, R., Bergstra, J., Goodfellow, I., Bergeron, A., Bouchard, N., Warde-Farley, D. and Bengio, Y., 2012. Theano: new features and speed improvements. arXiv preprint arXiv:1211.5590. Vancouver

Just want to put these picture because this project is currently lack of pictures.