Nonlinear Estimation and Modeling of fMRI Data using
Spatio-Temporal Support Vector Regression
Yongmei Michelle Wang, Robert T. Schultz, T. Todd Constable
and Lawrence H. Staib
Yale University
Information Processing in Medical Imaging, July 2003
(a)
(b)
(c)
(d)
Fig. 2. Simulated 2D + T data.
Top row: time T vs. spatial axis X; Bottom: spatial axis Y vs. X.
(a): Ground truth data;
(b): Simulated noisy data with noise level N(0, 30*30);
(c): Restored data by our SVR (W-model = 1);
(d): Gaussin smoothed data with s.t.d. = 0.5.
Fig. 3. Effects on time course with varying W-scale and W-model
in our SVR. (Simulated noise level: N(0, 30*30).)
Fig. 4. ROC curves for simulated noisy 2D + T data. (Average
effect of three noise levels.)
(a): by SVR (t > 7.8);
(b): by t-test (t > 4.2);
(c): by t-test (t > 2.3).
Fig. 5. Results comparison for real fMRI data from a
visuospatial task.
Fig. 6. Time courses of an activated voxel for
the real fMRI data in Fig. 5.
back to Home Page