Medical image fusion is becoming increasingly popular for enhancing diagnostic accuracy by intelligently 'fusing' information obtained from two different images. These images may be obtained from the same modality at different time instances or from multiple modalities recording complementary information. Due to the nature of the human body and also due to patient motion and breathing, there is a need for deformable registration algorithms in medical imaging. Typical non-parametric (deformable) registration algorithms such as the fluid-based, demons and curvature-based techniques are computationally intensive and have been demonstrated for mono-modality registrations only. We propose a fast and deformable algorithm using a 2-tiered strategy wherein a global MI-based affine registration is followed by a local piece-wise refinement. We have successfully tested this method on CT and PET images and validated the same using clinical experts.