Daily image-guided position check before each radiotherapy fraction
Every fraction of image-guided radiotherapy (IGRT) is judged against a single reference: the planning CT acquired at simulation, days before treatment begins. This dataset defines where the tumor is, how it moves, and what "correctly positioned" means for the entire course — often 20 to 40 fractions delivered over several weeks.
At simulation, the patient is positioned exactly as they will be for every treatment fraction, using custom immobilization: thermoplastic masks for head-and-neck, vacuum bags or alpha cradles for pelvis and thorax, knee/ankle wedges for lower limb reproducibility. A simulation CT (and often a 4D-CT for thoracic/abdominal sites, capturing the breathing cycle) is acquired, sometimes with IV contrast to delineate vasculature and soft-tissue targets.
Skin marks or tattoos are placed at the laser-defined origin, giving therapists an external reference for daily initial setup before any imaging occurs. This external alignment gets the patient within roughly a centimeter of the correct position — imaging then closes the remaining gap.
Oncologists and physicists contour a hierarchy of volumes per ICRU guidelines: the Gross Tumor Volume (GTV, visible disease), the Clinical Target Volume (CTV, GTV plus microscopic spread), and the Planning Target Volume (PTV, CTV plus a geometric margin accounting for setup uncertainty and internal motion).
That margin is not arbitrary. The van Herk formula, PTV margin ≈ 2.5Σ + 0.7σ, separates uncertainty into a systematic component Σ (the same error repeated every fraction, e.g. a consistent CT-to-treatment offset) and a random component σ (day-to-day scatter around that systematic error). Systematic errors are far more dangerous geometrically because they shift the entire dose distribution consistently — which is exactly what daily IGRT is designed to eliminate.
Because systematic error (Σ) is weighted roughly 3.5× more heavily than random error (σ) in the margin formula, correcting it via daily imaging — rather than merely averaging it out — is the single most effective lever for shrinking PTV margins.
The treatment isocenter — the point where the gantry, couch, and collimator axes intersect — is defined on the planning CT relative to the target. From this dataset, digitally reconstructed radiographs (DRRs) are generated: synthetic X-ray-like projections computed by ray-tracing through the CT volume along the treatment beam or imaging geometry.
These DRRs (or, for CBCT-based IGRT, the reconstructed planning CT volume itself) become the fixed reference against which every daily image is registered. Couch and gantry coordinates at simulation are logged as the baseline "zero" — stage one of the daily workflow is simply recalling this reference before any new image is acquired.
Before each fraction, the patient is imaged in the treatment position itself — not a separate room, not a separate day. Kilovoltage cone-beam CT (kV-CBCT), mounted directly on the linear accelerator gantry, has become the workhorse of daily IGRT because it produces a full volumetric dataset in about a minute, matched to the exact geometry the beam will use.
A kilovoltage X-ray source and a large flat-panel amorphous-silicon detector are mounted on retractable arms orthogonal to the treatment beam. As the gantry rotates, the source-detector pair sweeps around the patient — typically ~200° in "half-fan" offset-detector mode to cover a wide field of view with a smaller panel, or a full 360° in "full-fan" mode for smaller anatomy.
Hundreds of 2D projection radiographs acquired across the sweep are reconstructed into a 3D volume using filtered back-projection algorithms (classically Feldkamp-Davis-Kress, FDK), producing a CT-like dataset of the patient in the actual treatment position — mask on, couch top, immobilization device and all.
CBCT is not the only tool. Megavoltage portal imaging uses the treatment beam itself, detected by an electronic portal imaging device (EPID) — lower soft-tissue contrast but no separate imaging source, common for simpler weekly checks. Orthogonal kV planar pairs (e.g. ExacTrac) give fast 2D bony or fiducial-based alignment with less dose than a full CBCT.
Surface-guided optical systems (AlignRT, Catalyst) project structured light onto the skin to track body contour in real time with zero ionizing dose — valuable for frameless setups, breath-hold verification, and continuous intrafraction monitoring. Implanted electromagnetic transponders (Calypso) and radiopaque fiducial markers provide continuous or intermittent internal position feedback independent of external anatomy, particularly useful where soft-tissue targets (prostate, lung tumors) move relative to bone.
Every CBCT delivers imaging dose on top of the therapeutic dose — typically 10–30 mGy per scan depending on protocol, roughly comparable to a handful of conventional CT slices. Over a 40-fraction prostate course with daily CBCT, cumulative imaging dose can reach several hundred mGy, which must be justified against the geometric benefit gained.
ALARA (As Low As Reasonably Achievable) governs protocol selection: high-precision, high-consequence sites with significant day-to-day motion (prostate, lung SBRT, spine SBRT) justify daily CBCT, while simpler, more reproducible sites (whole breast, some head-and-neck) may use weekly kV or MV portal imaging instead, trading imaging frequency for lower cumulative dose where the anatomical stability allows it.
A single kV-CBCT scan (10–30 mGy) is small relative to a typical therapeutic fraction dose (1.8–2 Gy or more to the target), but summed daily over a full course it becomes a real, protocol-relevant dose that must be weighed against the PTV margin reduction it enables.
A reconstructed CBCT is only useful once it is compared to the reference. Registration software overlays today's volume onto the planning CT, computes the geometric transformation that best aligns them, and expresses the result as a shift the treatment couch must apply — the pivotal calculation step of the entire IGRT workflow.
Most daily IGRT uses rigid registration — assuming the patient's local anatomy moved and rotated as a solid body without deforming. Grey-value (intensity-based) algorithms, often using mutual information as a similarity metric, search over translations and rotations to maximize voxel-intensity agreement between the CBCT and the reference CT.
Clinicians choose a registration region-of-interest deliberately: bony anatomy matching (pelvis, spine, skull) is fast and robust but ignores soft-tissue organ motion; soft-tissue or organ matching (prostate gland, lung tumor, nodal volume) tracks the actual target but is more sensitive to image noise and requires visible contrast on CBCT. The automated result is always reviewed, and often adjusted, by the treating radiation therapist before it is applied.
A rigid-body position error has six independent components: three translations — lateral (left-right), longitudinal (superior-inferior), and vertical (anterior-posterior) — and three rotations — pitch, roll, and yaw. Traditional 3-axis couches can only correct the translations; any rotational error is left uncorrected, silently degrading target coverage.
Modern robotic 6D couches add pitch, roll, and yaw correction directly at the couch top, closing this gap. Registration software reports the full 6D vector, and the treatment console displays it as concrete numbers — for example lateral +4.2 mm, longitudinal −6.8 mm, vertical +3.5 mm, roll +0.6° — that the couch (or the therapist) then applies.
Not every measured shift triggers a couch move — imaging and registration themselves have noise, and moving the couch for sub-millimeter fluctuations wastes time without clinical benefit. Departments define an action level: a translational (and separately, rotational) threshold above which a correction is mandatory.
Common frameworks include the simple action-level protocol (correct any fraction exceeding a fixed threshold, typically 3–5 mm), the No Action Level (NAL) protocol (apply the mean systematic shift from the first few fractions to all subsequent ones without daily re-decision), and the extended NAL protocol (NAL combined with ongoing threshold-based correction for large individual-day deviations). The choice trades imaging/decision burden against residual systematic error.
A 3–5 mm action level is typical for conformal treatment of prostate and pelvic sites; SBRT protocols for lung, liver, or spine — where PTV margins are already only a few millimeters — often use tighter thresholds around 1–3 mm because the entire margin budget is smaller.
Once a shift exceeding the action level is confirmed, it must actually be applied. Modern linear accelerators pair with robotic six-degree-of-freedom couches capable of translating and rotating the patient with sub-millimeter precision, converting the registration software's output directly into physical motion.
Hexapod-style robotic couches (built on parallel-actuator platforms, conceptually related to Stewart platforms) sit atop the conventional couch and can execute all six corrections — three translations plus pitch, roll, and yaw — as a single combined motion rather than sequential axis-by-axis moves. The therapist reviews the calculated shift on the treatment console, confirms it against clinical judgment (checking for registration errors or gross anatomical changes), and triggers the correction.
Safety interlocks continuously monitor couch position against planned limits and patient clearance from the gantry and collimator during motion, halting immediately on any obstruction or unexpected deviation.
Because the initial registration and couch encoders both carry small uncertainties, many protocols acquire a second, verification image after the couch moves — either a repeat CBCT or a faster kV planar pair — to confirm the residual error has actually fallen within tolerance. If it has not (rare, but possible due to couch flex, patient shift during the move, or registration ambiguity), the correction-and-verify cycle repeats before treatment proceeds.
This closed-loop check is what allows departments to trust tight action levels: the system does not merely calculate a correction, it confirms the correction worked.
Couch correction addresses inter-fraction motion — the setup error that differs from one day's fraction to the next, arising from imperfect daily positioning, weight change, or bladder/rectal filling. It does nothing for intra-fraction motion — movement occurring during beam delivery itself, such as breathing, cardiac motion, peristalsis, or a patient shifting on the couch mid-treatment.
Intra-fraction motion requires separate solutions: respiratory gating (beam-on only during a defined breathing phase), breath-hold techniques, real-time tracking with electromagnetic transponders or optical surface monitoring that can pause the beam automatically, or continuous cine-MR imaging on MR-linac systems. A well-corrected setup at the start of a fraction can still be compromised by uncorrected intra-fraction drift.
Even after correction, a small residual random error persists (couch/imaging precision limits, typically well under 1–2 mm) — this residual σ still contributes to the van Herk margin formula, which is why daily IGRT reduces PTV margins substantially but rarely all the way to zero.
With the couch corrected and alignment confirmed, treatment finally proceeds — but the daily image acquired to verify setup carries a second, increasingly important use: it is also the earliest available evidence of how the patient's anatomy is changing over the course of treatment, feeding directly into adaptive replanning.
Once the verification image confirms the residual error is within the action level, the treatment console cross-checks the plan parameters — monitor units, beam energy, field shape and gantry/collimator angles — against the prescribed plan before allowing the beam to be enabled, an independent safety check layered on top of the geometric one. Only then does dose delivery begin, now aimed at a target whose position matches the planning CT to within a millimeter or two rather than the several-millimeter uncertainty of setup-marks-only positioning.
Returning to the van Herk formula, PTV margin ≈ 2.5Σ + 0.7σ: daily image guidance drives the systematic component Σ toward zero, because any consistent offset is caught and corrected before it can repeat fraction after fraction. Only the smaller random component σ remains, dominated by residual registration and couch precision rather than gross setup error.
The clinical payoff is a shrinking PTV margin — commonly from 7–10 mm with setup-marks-only positioning down to 3–5 mm with daily volumetric IGRT. Smaller margins mean less healthy tissue irradiated, which either reduces toxicity at a fixed prescription dose or permits dose escalation and hypofractionation (fewer, larger fractions) — the geometric precision underlying modern SBRT regimens.
Daily CBCT does more than catch positioning error — it reveals real anatomical change over the treatment course: a shrinking tumor, a bladder or rectum filling differently than at simulation, weight loss altering the whole body contour, or a resolving surgical seroma. When these changes are large enough to compromise target coverage or organ-at-risk sparing, the plan itself needs updating, not just the couch position.
Online adaptive radiotherapy platforms (such as Ethos, and MR-linac systems with integrated cine-MRI) now use that day's imaging to re-contour and re-optimize the treatment plan on the couch, before treating — turning the daily verification image from a passive quality check into an active input for a plan that adapts to the patient's anatomy each and every fraction.
Online adaptive radiotherapy closes the final loop: instead of only asking "is the patient where the plan expects?", the system asks "does the plan still fit the patient today?" — and rebuilds it in minutes if it does not.