P1: Learning from fMRI Brain Imaging
The project concerns with predicting the users’ action giving fMRI brain images (approximately 5,000 voxels) that on the “Region of Interest” (ROIs). In the experiment, each subject performed a total of 54 trails. In some trails, the subject simply rested, or gazed at a fixation point on the screen. For other trails (about 40), the image and sentence were shown, and the subject decided whether the sentence correctly describe the image or not. For the first 4 seconds, first stimulus was displayed (either sentence or image). For the next 4 seconds, first stimulus was replaced with blank screen. Then, the second stimulus was presented for the next 4 seconds or until the subject pressed a mouse button. If the first stimulus was an image, then the second stimulus was a sentence (and vice versa). Finally, the subject had a 15-seconds resting period where the second stimulus was removed. Image data of the subjects are collected once every 500msec for a total of 27 seconds (54 image total). The main focus of this project is to predict when the subject was reading a sentence or perceiving a picture, using the fMRI data of each subject’s brain. Main challenges are finding useful features from 5,000+ sensors data, developing machine learning model to predict the binary classification, and whether looking at image/sentence in different order has any impact or not.
September 30 – Learn and understand the problems and the data. Programming Language Selected.
October 31 – Choosing features and classifications. Finish code.
November 13 – Finalize and conclude result
November 20 – Presentation Slide
November 27 – Final Paper
Mitchell et al, 2004: http://www.cs.cmu.edu/~tom/mlj04-final-published.pdfJiayu Zhou, Jun Liu, Vaibhav A. Narayan, Jieping Ye: Modeling disease progression via fused sparse group lasso. KDD 2012: 1095-1103