Breathing moves the target: 4DCT, gating, and proving the chain works with a motion phantom
Why respiratory motion distorts CT geometry and blurs emission images, how phase and amplitude binning differ, where external surrogates mislead you, and an end-to-end validation you can run with a 4DCT respiratory motion phantom.
A lesion in the lower lobe of the lung travels a substantial distance between end-inhale and end-exhale, and one near the dome of the liver travels further. The scanner does not know this. It samples one slab of anatomy at one moment and the next a second later, then reconstructs the two as though they belonged to the same instant. What comes out is not a blurred image. It is a geometrically wrong one.
Departments buying 4DCT, gated SPECT or PET, or planning radiotherapy in the thorax and upper abdomen, are buying a chain: a surrogate sensor, an acquisition mode, a sorting algorithm, a planning system and, in radiotherapy, a delivery system that has to respond in time. Every link has a failure mode that produces a plausible-looking image. A motion phantom is the only practical way to find out which link is lying.
What motion actually does
On CT, motion distorts geometry rather than merely blurring it. A spherical lesion can reconstruct as a disc, a banana or two separate objects depending on where in the cycle the couch happened to pass. The reconstructed volume can be too large or too small, and the image gives no clue which way the error went.
On emission imaging, motion trades concentration for volume. A SPECT or PET acquisition lasts many breathing cycles, so counts from a moving lesion smear along the path. The lesion looks larger and colder than it is, and a measurement made without motion management cannot be compared with a gated one as a trend.
On fused images, motion breaks registration. If the CT is a snapshot of one phase and the emission scan an average over many, they do not align. At the lung and liver boundary this produces the familiar attenuation correction artefact, a cold band following the diaphragm rather than any anatomy.
On radiotherapy plans, motion turns a margin into a guess. Contouring on a single free-breathing scan and adding an isotropic margin guesses at both the size and the direction of the excursion. The purpose of 4DCT is to replace that guess with a measured envelope.
How a 4DCT dataset is actually built
The scanner oversamples: at each couch position it acquires for at least one full breathing period plus the reconstruction window, in cine mode or at very low pitch, while a surrogate signal is recorded alongside. Nothing is gated during acquisition. The sorting happens retrospectively, which is why a 4DCT costs dose and why the sort is where the errors live. Projections are assigned to bins, conventionally ten, labelled from end-inhale through end-exhale and back. Two sorting methods are in general use, and they fail differently.
Method
Sorts by
Handles well
Fails when
Phase binning
Fraction of time through each detected cycle
Shallow or shifting breaths, since every cycle feeds every bin
Amplitude varies between cycles, so a 30 per cent bin from a deep breath sits at a different position from one from a shallow breath
Amplitude binning
Surrogate displacement value
Varying breath depth, because position maps to position
The patient misses a given amplitude in some cycles, leaving gaps that have to be interpolated
From the sorted phases come the derived series, and confusing what each one is for is a common and consequential error.
Series
Built from
Use it for
Do not use it for
Individual phase bins
Sorted projections
Measuring excursion, contouring the target at each phase
Dose calculation from a single phase alone
MIP, maximum intensity projection
Brightest voxel across phases
Drawing the internal target volume for a bright lesion in aerated lung
Targets against soft tissue of similar density, where it is misleading
AIP, average intensity projection
Mean voxel across phases
The density dataset for dose calculation
Measuring excursion or defining the target
The visible sign that a sort has gone wrong is a stair-step at the diaphragm, or a structure appearing twice or cut in half at a couch position boundary. The honest response is usually to coach the patient and rescan rather than reprocess. Regular breathing fixes more artefacts than software does.
Gating in practice
End-exhale is the usual window because it is the most reproducible part of the cycle. Patients sit at the bottom of the curve longer than at the top, so the position is easier to hit repeatedly.
Duty cycle trades against residual motion. A narrow window leaves less residual motion inside it and lengthens the treatment or acquisition. The compromise commonly quoted around the exhale portion is a starting point, not an answer. Set it from the patient's recorded trace.
Measure before deciding to gate at all. Many patients move less than the department assumes. Where the trace and the 4DCT show small excursion, an internal target volume with sensible margins beats gating someone who cannot breathe regularly. Breath-hold and abdominal compression suit particular patients too, and gating is not automatically the best of them.
Coaching changes outcomes. Audio prompting, a visual display or a technologist counting with the patient improves regularity enough to change which sorting method works. Whatever coaching you use clinically must be used during phantom testing too, or you have not tested the clinical chain.
The surrogate problem
The gating hardware watches a marker block on the abdomen, a pressure belt, or an optical surface map. The target is inside the patient. Everything downstream rests on an assumption that the relationship between the two is stable, linear and in phase. It is regularly none of the three.
Hysteresis: the internal path on inhalation is not the path on exhalation, so one surrogate value maps to two internal positions.
Phase lag: the abdominal surface and the diaphragm do not necessarily peak together, and the offset varies between patients.
Baseline drift: over a fifteen or twenty minute session the patient relaxes and the whole trace shifts, moving the internal position while the window stays where it was set.
Marker placement: a block placed a few centimetres from where it sat at planning changes the reported amplitude without changing anything inside the patient. Photograph it and mark the skin.
So verify the correlation per patient where the tools allow it, re-verify if the session runs long, and set margins that survive a surrogate being wrong by a quantity you have measured rather than assumed.
What a motion phantom is for
A respiratory motion phantom drives a target of known size, shape and density along a reproducible, programmable trajectory while presenting a surrogate signal to your gating hardware. Because you know what the target did, you can compare it against what the system says it did. Nothing else gives you that.
The Quattro 4DCT respiratory motion phantom from Leeds Test Objects is the unit Vvon supplies for this work: an anthropomorphic thorax body with lung, soft tissue and spine equivalent sections and moving inserts, driven by a separate programmable motor unit. Ask for the current motion range, waveform library and insert options when you specify it rather than working from figures in an article. Configuration matters more than the headline, and the interface to your particular gating system matters most of all.
Commissioning the sort. Image a target of known dimensions moving at a known amplitude and measure its length in each phase bin, on the MIP and on the AIP. If the MIP extent is not close to the target length plus the set amplitude, your sorting or your synchronisation is wrong, and you have found it on a phantom rather than a patient.
Measuring latency in millimetres. Every gating chain delays between the surrogate crossing a threshold and the acquisition or beam responding. In milliseconds that means nothing to a clinician. Run the phantom at a known velocity and the same delay becomes a displacement in millimetres, which is the number that goes into a margin.
Testing irregular breathing. Run a patient-like trace with varying amplitude and a drifting baseline. A sine wave passes everything, which is why vendors ship one.
Emission gating. With a fillable insert, measure recovered activity concentration gated against ungated for the same target, and find out whether your gated numbers are quantitative or merely different.
Annual constancy. Repeat the identical test against the commissioning result. A phantom that yields a number you can compare year on year is doing its job; one that yields a pass is not.
The predictable mistake is buying the phantom, running the vendor's sine wave once at acceptance, filing the printout and shelving the box. If the phantom has not been out since the last software upgrade, nobody knows whether the upgrade changed the sort.
An end-to-end test you can actually run
Set the phantom on the couch with the target insert at a known amplitude and period, align to the room lasers and record the setup position.
Acquire a 4DCT using the clinical protocol, with the coaching you would give a patient, then sort and reconstruct every phase plus MIP and AIP.
Contour the target on each phase and on the MIP. Compare measured excursion against the set amplitude, and measured volume against the known volume. Write down both differences.
Plan on the AIP using the internal target volume derived from the MIP, with your clinical margins.
Deliver to the moving phantom with film or a detector array in the target plane, gate on and then off.
Analyse against the calculated dose using gamma criteria your department agreed beforehand and that appear in the report. Criteria chosen after seeing the result are not criteria. Then repeat the whole run with a deliberately irregular trace.
File it with the date, the phantom serial number and the software version of every component in the chain. Version numbers are what make a repeat test meaningful after an upgrade, and they are what is always missing three years later.
What to check before you specify one
Interface compatibility with your gating system. Confirm in writing, naming both products, that your equipment can read the phantom's surrogate output. A brochure photograph is not an interface specification.
Programmable waveforms including recorded patient traces, not sinusoids alone.
Inserts matched to your work: a fillable insert if you gate emission imaging, a film or detector cavity for radiotherapy end-to-end tests, tissue equivalent sections if you calculate dose on the images.
An independent way to verify amplitude, plus calibration and traceability documentation and a clear answer on who recalibrates the drive unit.
Somewhere dry to keep it. A plastic body and an electronic drive unit left in an uncontrolled room through a Dhaka monsoon will warp, corrode, or both. Put the drive unit on conditioned power: supply transients kill motor controllers quietly.
Motion management is still uncommon in Bangladesh, which cuts two ways. A department that has it holds a real clinical advantage. A department that has it and has never validated it is running margins on a vendor default, with no way of knowing which of its lower-lobe measurements can be believed.