Machine Learning Approaches to fMRI

Tom Mitchell
Carnegie Mellon University
AI and Learning

Machine learning algorithms such as classifier learning methods are being used
increasingly for analyzing fMRI data. Whereas many approaches to fMRI analysis seek
to answer the question "which brain regions are active on average during cognitive task
X?" classifier learning methods can be used to answer questions such as "is the subject
performing cognitive task X or Y during this particular time interval?" The crossvalidated
accuracy of classifier predictions provides an objective, assumption-free basis
for assessing the validity of learned models.

This lecture will cover classifier learning methods such as Support Vector Machines
(SVMs) and Bayesian classifiers, and their use for analyzing fMRI data. We will cover
several examples, such as training classifiers to distinguish whether a human subject is
comprehending a sentence or comprehending a picture, and whether they are reading a
word about tools or buildings. Trained classifiers for these tasks achieve prediction
accuracies of 80% to 95% for new data taken from the same subject, and in some cases
can be used for predictions on new subjects outside the training set. We will examine
feature selection methods crucial to achieving these accuracies, and consider the role of
these classifiers in studying human cognitive processing. We will also explore related
time series modeling methods based on temporal Bayesian networks, and their potential
role in modeling cognitive processes.

Presentation (PowerPoint File)

Back to Graduate Summer School: Mathematics in Brain Imaging