A round burgundy crest emblem with simple gold ring detailing, plain and formal in style
A wide horizontal photograph of mountain scenery near Banff with pale turquoise sky, dark evergreen slopes, and a soft muted mood
A round burgundy crest emblem mirroring the first, with plain gold ring detailing and a formal quiet mood

Annual Scientific Meeting of the Canadian Organization of Medical Physicists — July 9–12, 2014, The Banff Centre, Banff, Alberta, Canada

Deformable Image Registration For Adaptive Replanning

Radiotherapy is planned from an image acquired at a particular moment, while the patient’s anatomy changes throughout treatment. Tumours may shrink, organs can fill or empty, and weight loss may alter the relationship between the target and nearby organs at risk. Deformable image registration (DIR) provides a way to estimate how anatomy has moved or changed between image sets, supporting more responsive treatment planning.

For medical physicists, the subject sits at the intersection of image guidance, treatment adaptation, contour propagation and dose calculation. The archived COMP 2014 archive offers useful professional context for studying how these technical discussions fit within a scientific meeting, where clinical implementation, quality assurance and collaboration are considered alongside research results.

Why Deformable Registration Matters

Rigid registration applies a single translation and rotation to align two image sets. This approach can be effective for bony anatomy, such as matching a planning CT with a cone-beam CT using the spine or pelvis. It becomes less reliable when soft tissue deforms, organs change volume, or a treatment site shifts internally.

DIR estimates a spatial transformation that varies across the image. A voxel in the planning scan is mapped to a corresponding location in a later scan, creating a deformation field. That field can help transfer structures, compare anatomy and estimate how the delivered dose relates to the patient’s current geometry.

The word “estimate” is important. A deformation field is generated by an algorithm, not observed directly throughout the body. Registration quality depends on image contrast, artefacts, field of view, respiratory motion, patient positioning and the assumptions built into the software. A visually convincing overlay can still contain clinically significant local errors.

From Image Alignment To Clinical Meaning

Adaptive replanning usually begins with an indication that the original plan may no longer provide the intended balance between target coverage and normal-tissue protection. In head and neck treatment, progressive weight loss or tumour response can change dose distribution. In pelvic treatment, bladder and rectal filling may affect anatomy from fraction to fraction. In lung treatment, respiratory phase and tumour motion complicate interpretation.

A DIR workflow can propagate contours from a planning CT to an online or repeat image, giving the team a starting point for review. It can also support accumulation of dose across fractions, although dose warping and accumulation require careful attention to coordinate systems, tissue changes and uncertainty. Propagated contours must be checked by an appropriately trained clinician rather than accepted as final simply because the software reports successful registration.

The clinical question should lead the process. A useful review asks whether the anatomy has changed enough to affect target dose, whether an organ at risk has moved into a high-dose region, and whether a new plan would materially improve treatment. Registration is a tool for answering those questions, not a substitute for clinical judgement.

Checking The Deformation Field

Validation should combine quantitative measures with expert visual assessment. Dice similarity coefficient, Hausdorff distance and landmark displacement can describe agreement between structures, but each metric has limitations. A high overlap score may hide a clinically important error at the edge of a target, while a modest score may reflect a genuine anatomical change rather than algorithm failure.

Landmark checks are especially useful when identifiable points are available, such as vessel bifurcations, surgical clips or vertebral features. Difference images, checkerboard displays and contour overlays can reveal stretching, folding or implausible motion. Jacobian determinant maps may also highlight local expansion and contraction, although their interpretation depends on image quality and the registration model.

A robust programme records when DIR is appropriate, which image pairs may be registered, who reviews the result and what action follows an unacceptable output. Commissioning should include representative patient anatomy rather than relying solely on vendor test cases. Software upgrades, new scanners and altered reconstruction protocols should trigger a proportionate review because they can change registration performance.

Applying DIR In Australian Practice

Australian departments may face very different operating conditions. A metropolitan service in Sydney, Melbourne or Brisbane may have access to multiple scanners, dedicated adaptive radiotherapy staff and research support, while a regional centre may work with fewer specialists and longer referral pathways. The workflow must be safe when a physicist is managing several responsibilities and an oncologist is reviewing images between clinics.

Long travel distances also matter. A patient from regional Western Australia or northern Queensland may not be able to return promptly for repeated planning, so adaptation criteria should be clinically meaningful and communicated clearly. Remote review can assist collaboration, but image transfer, latency, cybersecurity and state-based privacy requirements need to be addressed before a service relies on it.

Local professional language and expectations shape implementation as well. A department might describe a short afternoon task as an “arvo” review, yet the casual wording should not obscure formal sign-off requirements. Australian public hospitals and private providers also operate under different procurement, staffing and governance arrangements. Engagement with the Australasian College of Physical Scientists and Engineers in Medicine, local radiation regulators and the department’s quality system can help align DIR use with existing responsibilities.

Practical Review Points For A DIR Workflow

Before clinical use, confirm the following:

  • Image orientation, voxel spacing, field of view and acquisition date are correct.
  • The registration region excludes irrelevant anatomy and includes the structures needed for review.
  • Rigid alignment has been assessed before any deformable transformation is applied.
  • The deformation field contains no obvious folding, discontinuity or implausible displacement.
  • Propagated contours have been reviewed at clinically relevant slices and planes.
  • Dose comparison uses appropriate recalculation, density handling and uncertainty assessment.

When a patient appears suitable for adaptation, document the evidence:

  • The anatomical change and the image on which it was observed.
  • The target and organ-at-risk structures affected by that change.
  • The original plan’s limitations under the current anatomy.
  • The proposed adaptation, including any new scan or contouring requirements.
  • Independent checks completed by the physicist, radiation therapist and radiation oncologist.
  • The decision, reviewer names and date in the clinical record.

Building A Sustainable Adaptive Service

Adaptive replanning is a service, not a single software feature. It involves image acquisition, registration, contour review, treatment planning, plan checking, communication and delivery. Each stage needs an owner and a defined turnaround time. If the workflow adds hours to a busy treatment day, the department should decide in advance which cases receive priority and how urgent decisions are escalated.

Automation can reduce repetitive work, especially for image matching and contour propagation, but automation also creates a risk of unnoticed error at scale. Human review should focus on regions where a small geometric difference has a large dosimetric consequence. For example, a few millimetres near the spinal cord, optic apparatus or bowel may matter more than a larger difference in low-dose tissue.

Training should use cases that reflect local practice, including head and neck, prostate, cervix, lung and stereotactic treatments where relevant. A short teaching session can cover registration algorithms, similarity metrics, failure modes and dose accumulation without turning the subject into a purely mathematical exercise. The archived conference material and registration details can also help place technical learning within the wider culture of professional meetings, examinations and peer exchange.

A sensible first implementation may be retrospective testing followed by observer-only clinical use. The department can compare algorithmic contours with expert contours, measure review time, collect failure examples and establish escalation rules. Once the evidence is satisfactory, prospective use can begin for a narrow patient group before the service expands.

DIR has genuine potential to make radiotherapy more responsive to anatomical change, but its value depends on disciplined interpretation. Registration accuracy, contour reliability and dose assessment must remain visible to the clinical team. With clear governance and practical training, Australian services can use deformable mapping as a measured step towards safer, more personalised adaptive treatment.

Explore the conference archive, discuss the workflow with colleagues and develop a small validation set drawn from your own scanners and patient population. A carefully documented pilot can turn deformable image registration from an abstract research topic into a dependable part of everyday clinical decision-making.