Dear colleagues,
We will have a seminar on next wednesday, 12th March about medical image analysis by Dr. Lee Harrison, UCL. We would like to invite you and your group members to join the seminar. The detail of the seminar is as follows. Please could you circulate this email to your group members, and we welcome every one to join the seminar.
Best regards,
Fan Zhang Computer Vision and Pattern Recognition Group, Computer Science Department, University of York, UK
-----------------------------------------------------------------------------------------------------------------------------
Lee Harrison (University College London) Title: "Spatial models of functional magnetic resonance images"
11:15AM - 12:30AM, 12th March, Wednesday in Room CS103, Computer Science Building -----------------------------------------------------------------------------------------------------------------------------
"I will talk about Bayesian spatial models of fMRI time-series data. In particular, we use a weighted graph Laplacian (WGL) to partition a brain volume into computationally manageable segments, using an isoperimetric partitioning algorithm (Grady 2006). Data from each segment is then entered into a 2-level general linear model (GLM) with matrix-variate normal densities over likelihood and prior at the 1st and 2nd levels. The spatial covariance matrix of the prior over voxels is given by the heat kernel of a WGL and all parameters and hyperparameters e.g. controlling dispersion of the heat kernel, are optimized using an EM algorithm with Fisher-Scoring ascent scheme. The benefits of this approach are that GLM parameters have an explicit and (generally) non-stationary spatial model. This allows formal model comparison of different hypotheses about how data are generated, e.g. from a stationary or non-stationary spatial process."