Multimodality Image Registration and Fusion
Registration concepts
Understand rigid vs deformable registration. Rigid (translation + rotation, sometimes scale) preserves distances; deformable allows local warping. For most RT head/pelvis workflows, rigid registration to CT is standard; deformable is used where anatomy changes (e.g., adaptive planning) but must be validated.
Registration methods
Point-based (fiducials), surface/landmark, and intensity-based (mutual information for CT-MR since intensities differ). Mutual information is the workhorse for CT-MR fusion. Understand resampling, interpolation, and the reference/fixed vs moving image roles.
Accuracy and validation
Assess registration with anatomical landmarks, Dice/overlap for contours, and visual checks in multiple planes. Errors arise from patient positioning differences, distortion, and interpolation. Always verify target alignment before propagating contours.
Traps
Questions may ask which image is "fixed" - the reference dataset does not change; the moving image is transformed. Do not assume deformable registration is always more accurate; it can introduce non-physical warps. CT-MR fusion requires accounting for MR geometric distortion.
Rule of thumb
Register MR to the planning CT (CT is the dose-calculation reference), then verify with at least three landmarks in three planes before contour propagation.
Worked example
For a brain tumor, rigidly register the T1 post-contrast MR to the planning CT using mutual information, then confirm alignment at the ventricles, falx, and skull base before transferring GTV contours.