Figure 3. Computing a training set.

The PET images F_i (where i is in the set of training subjects) are deformably registered onto the PET template F̄. The structural images S_i are first co-registered with their corresponding PET images, followed by a deformable registration onto the structural image template S̄. The affine transformation ${\mathit{T}}_i^{\prime }$ preceding the deformation is constrained to be the affine transformation obtained from PET-to-PET template registration.