Fig. 2.

Discrimination map or support vector machine (SVM) weight vector: continuous versus episodic course (top), continuous course versus healthy individuals (bottom). The colours represent the weight of each voxel in the classification function (the red scale represents positive weights and the blue scale represents negative weights). The SVM weight vector is a linear combination or weighted average of the support vectors, that is the training examples that are most difficult to separate and define the decision boundary. The weight vector is therefore a spatial representation of the decision boundary. Every voxel contributes with a certain weight to the decision boundary or classification function. Given a positive and a negative class (e.g. +1=episodic group; −1=continuous group), a positive weight for a voxel means the weighted average in that voxel was higher for the episodic group, and a negative weight means the weighted average was higher for the continuous group. Because the classifier is multivariate by nature, the combination of all voxels as a whole is identified as a global spatial pattern by which the groups differ (the discriminating pattern). Therefore, the discrimination map should not be interpreted as a standard statistical parametric map resulting from a mass-univariate statistical test to find group differences, and no local inferences should be made based on the SVM weights.