B0257
Title: A Bayesian non-parametric Potts model for fMRI presurgical planning
Authors: Timothy Johnson - University of Michigan (United States) [presenting]
Abstract: There is growing interest in using fMRI data in clinical practice, especially for presurgical planning. A fully Bayesian model is presented for fMRI data that may be more suitable in clinical applications than standard fMRI tools. A time-varying autoregressive model is used to capture non-stationary behaviour over time. Low-frequency drift is modelled using adaptive B-spline bases. A conditional autoregressive-type prior is placed on the model variances. Hyperpriors are specified on the HRF parameters allowing greater modeling flexibility on the shape of the HRF. A non-parametric Potts model is used to partition the parameters of interest into deactivated, activated, and null classes. The modelling approach is compared to a standard mass-univariate approach on a presurgical fMRI dataset in which the patient has a temporal lobe glioblastoma.
