Imagine trying to hit a moving target with millimetre precision, while the target moves unpredictably with every breath the patient takes. That is the daily reality of lung radiotherapy.
Lung tumours do not sit still. With every respiratory cycle, they shift in three-dimensional space - anteroposteriorly, supero-inferiorly, and laterally. The extent of this motion is highly variable: for some patients, a tumour may move less than 2mm; for others, it can travel more than 3cm per breath, particularly for lower lobe lesions close to the diaphragm.
This is not a minor calibration issue. Respiratory motion is one of the most significant sources of geometric uncertainty in thoracic radiotherapy, and it has been a defining problem in the field for decades. Managing it well is the difference between precise, lung-sparing treatment and an unnecessarily large, toxicity-prone plan.
When respiratory motion is not properly accounted for in radiotherapy planning, three classes of problems emerge, each with real consequences for patient outcomes.
A conventional 3DCT scanner captures a single snapshot of thoracic anatomy. If that snapshot coincides with any phase of the breathing cycle arbitrarily - inhale, exhale, or mid-breath, the resulting image is unrepresentative of the tumour's true average position. Motion artefacts distort the apparent shape and size of the tumour, and the resulting GTV contour may be inaccurate before any margin is even applied.
To compensate for motion uncertainty, planners historically added large, arbitrary margins to the Clinical Target Volume (CTV) to create the Internal Target Volume (ITV). For a lower lobe tumour, this could mean a 20mm craniocaudal margin, a substantial expansion that envelops significant volumes of normal lung. More lung in the high-dose region means greater risk of radiation pneumonitis and long-term pulmonary fibrosis.
Even with a well-constructed plan, intra-fraction motion means the dose is not deposited where the plan assumed. Each breath cycle smears the dose distribution, blurring the sharp edges that conformal techniques like VMAT and SBRT depend upon for their tissue-sparing advantage.
Before diving into motion management techniques, it helps to be precise about the volume hierarchy we are working within. This framework, defined by the ICRU (International Commission on Radiation Units and Measurements), provides the conceptual scaffolding for all radiotherapy planning decisions.
The macroscopic extent of disease as visible on imaging CT, PET-CT, or MRI. This is what we can see. Contouring accuracy here depends heavily on image quality and, critically, on whether motion has distorted the apparent tumour shape.
GTV plus a margin to account for subclinical microscopic disease spread not visible on imaging. For early-stage lung SBRT, the CTV is often equivalent to the GTV, since the biological rationale for microscopic margins is different in the hypofractionated setting.
This is where motion management becomes decisive. The ITV is the CTV plus an internal margin (IM) that accounts for variations in the CTV's position, size, and shape due to physiological factors, predominantly respiration. The quality of the ITV directly determines how much normal lung will ultimately receive high dose. A poorly estimated ITV is either too large (toxicity risk) or too small (geographic miss risk).
ITV plus a setup margin (SM) for day-to-day patient positioning uncertainties. In our study, we applied a uniform 3mm margin to all ITVs to create the final PTV, isolating the ITV differences as the primary variable of interest.
Motion management in radiotherapy broadly divides into four philosophical approaches: encompassing the motion, reducing the motion, gating the beam to a specific phase, or tracking the tumour in real time. Each has its clinical niche.
Techniques like 4DCT and slow CT capture the full range of tumour excursion and plan to treat across it. The ITV envelops all positions. High confidence in tumour coverage; tradeoff is larger volumes.
Deep Inspiration Breath Hold (DIBH) and Active Breathing Control (ABC) freeze the tumour at a reproducible respiratory phase. Effective for reducing superior-inferior motion, especially in breast and lung.
Beam is activated only during a specific window of the breathing cycle, either amplitude-based or phase-based. Extends treatment time but dramatically reduces the volume that needs to be irradiated.
CyberKnife with fiducial markers, or electromagnetic transponders, allow the beam aperture to follow tumour motion continuously. Maximum accuracy; requires significant infrastructure and implant procedures.
In clinical practice at most centres including Amrita Institute of Medical Sciences where my undergraduate research was conducted - the choice between these approaches depends on patient anatomy, tumour location, institutional technology, and patient compliance. For the majority of lung SBRT cases, 4DCT-based motion encompassing remains the workhorse technique.
Conventional CT acquires a three-dimensional anatomical snapshot. 4DCT adds a fourth dimension: time. By acquiring CT data continuously across multiple respiratory phases at each couch position, 4DCT generates a series of 3D image sets, typically 10 phases spanning from peak inhalation (0%) to peak exhalation (50%) and back, each representing the patient's anatomy at a distinct moment in the breathing cycle.
During a 4DCT acquisition, the scanner uses a cine mode rotating continuously at each table position rather than advancing steadily through the patient. A respiratory surrogate signal (either an external device like the Varian RPM block or the Anzai pressure belt, or an internal surrogate via GE's Deviceless 4D system) tracks the breathing cycle in real time. Each acquired image slice is tagged to its corresponding respiratory phase based on where it falls in the breathing waveform.
Once all data is collected, images are retrospectively sorted, or "binned", into their respective phases, producing a complete 4D dataset. From this, MIP, Average Intensity Projection (AveIP), and individual phase reconstructions can be generated for planning and contouring.
A notable development in 4DCT technology is the GE Deviceless 4D (4D-DL) system, which eliminates the need for external respiratory tracking devices entirely. Instead of measuring chest wall motion with a pressure belt or infrared block, 4D-DL uses internal anatomical features, lung area, body area, lung density, air content, lung extension, and lung-body proportion extracted directly from the CT images themselves using CT numbers.
In our study at AIMS Kochi, we used a 16-slice GE Discovery 580 WRT scanner with GE Advantage 4D software, acquiring 4DCT images in free breathing and reconstructing 10 respiratory phases plus MIP images, all at 2.5mm slice thickness for three patients with lung SBRT planned at upper, middle, and lower lobe locations respectively.
Once 4DCT data is acquired, the Maximum Intensity Projection (MIP) is the primary tool for ITV delineation. Understanding how it works explains why it is so well-suited to this task.
The MIP algorithm renders a composite image in which each voxel is assigned the highest density value (highest CT number) encountered across all respiratory phases at that spatial location. For a solid lung tumour which has high density relative to the surrounding aerated lung parenchyma this means that every position the tumour occupied throughout the breathing cycle is represented in the MIP image as a high-intensity region.
The result is a single image in which the tumour's maximum excursion envelope is clearly visible. Contouring the GTV on the MIP produces an ITV that, by definition, encompasses all positions the tumour occupied during the acquisition period. No arbitrary margins are needed - the motion is encoded directly into the dataset.
For tumours adjacent to high-density mediastinal structures, the MIP can overestimate the ITV due to superimposition of the mediastinum. In these cases, individual phase contouring and Boolean union operations provide a more accurate ITV. This is a known limitation and an area where AI-assisted auto-contouring which is precisely what my VINESeg project addresses holds significant clinical promise.
My undergraduate thesis, conducted at AIMS Kochi under the supervision of Dr. Haridas M Nair (Department of Radiation Oncology), directly compared PTV volumes generated from three imaging approaches across three lung SBRT patients, one each with upper lobe, middle lobe, and lower lobe tumours.
All three patients were immobilised supine in VacLoc with arms by the side and scanned on a 16-slice GE Discovery 580 WRT CT simulator. For each patient, three separate simulation approaches were performed and three corresponding PTVs generated:
A uniform 3mm PTV margin was applied to all three ITVs to account for setup uncertainties, ensuring the PTV comparison reflected only differences in ITV estimation methodology.
| Patient | Location | GTV (cc) | PTV1 Free Breath (cc) | PTV2 Inhale/Exhale (cc) | PTV3 4D MIP (cc) | Reduction vs PTV1 |
|---|---|---|---|---|---|---|
| Patient 1 | Upper Lobe | 9.05 | 44.9 | 23.2 | 24.5 | 45.5% (MIP) |
| Patient 2 | Middle Lobe | 28.6 | 155.9 | 58.2 | 63.0 | 59.6% (MIP) |
| Patient 3 | Lower Lobe | 6.7 | 109.7 | 32.9 | 25.8 | 76.5% (MIP) |
| Average PTV reduction (4DCT MIP vs Free Breathing 3DCT) | ~60% | |||||
The magnitude of PTV reduction from 4DCT was not uniform, it correlated directly with expected tumour motion amplitude. The lower lobe patient showed a 76.5% reduction, the middle lobe 59.6%, and the upper lobe 45.5%. This confirms the anatomical intuition: lower lobe tumours are most affected by diaphragmatic excursion and therefore gain most from precise motion characterisation.
On reviewing the CT images, an important qualitative observation emerged: the 4DCT MIP accounted for tumour motion in the anteroposterior direction more effectively than the inhale/exhale breath hold technique, which predominantly captured supero-inferior motion. This is clinically significant, it means breath hold techniques, while valuable, may underestimate the true motion envelope in certain planes, and 4DCT provides a more complete three-dimensional picture.
For the lower lobe patient, this resulted in the 4DCT MIP producing a smaller PTV (25.8cc) than the inhale/exhale method (32.9cc), despite encompassing a more complete motion envelope demonstrating that arbitrary margins significantly overestimate the ITV even for breath hold techniques.
The 3DCT free breathing approach relies on location-based rules of thumb: 5mm for upper lobe, 10mm for middle lobe, 20mm craniocaudal for lower lobe. For Patient 3 (lower lobe, GTV just 6.7cc), this produced a PTV of 109.7cc more than 16 times the size of the visible tumour. The 4DCT MIP produced a PTV of 25.8cc for the same patient. The clinical implication of that difference in terms of normal lung irradiation and toxicity risk is profound.
4DCT simulation provides an individualised approach to lung SBRT planning that correlates motion management directly with each patient's respiratory cycle. By generating reconstructed MIP and Average Projection images, it enables more accurate ITV estimation and consistently reduces the volume of normal lung irradiated, without compromising confidence in tumour coverage. The safe balance between effective tumour control and normal tissue toxicity is not just achievable with 4DCT; it is substantially more achievable than with free-breathing 3DCT alone.
The challenges exposed by respiratory motion management in lung radiotherapy are becoming increasingly tractable thanks to two converging technological trajectories: AI-powered auto-contouring and online adaptive radiotherapy.
Manual GTV and ITV delineation on 4DCT is time-consuming and subject to inter-observer variability. Deep learning models particularly those built on architectures like MONAI and trained on datasets such as LIDC-IDRI are increasingly capable of segmenting lung nodules and tumours across multiple CT phases with accuracy that approaches or matches that of expert radiologists. Automating this process has the potential to dramatically reduce planning time, increase consistency, and make 4DCT-based personalised motion management accessible at higher volume and in resource-limited settings.
This is precisely the clinical problem I am working to address with VINESeg a MONAI-based lung nodule auto-contouring pipeline trained on LIDC-IDRI data, designed to produce DICOM RT-Structure outputs that integrate directly into clinical treatment planning workflows.
Systems like the MR-Linac take motion management into a new paradigm entirely. By combining real-time MRI imaging with radiation delivery, they allow the treatment plan to be adapted online every session based on the patient's anatomy as it exists at that moment. This eliminates the requirement for population-based or even patient-specific pre-treatment motion models; the treatment adapts to reality, not to a CT acquired weeks earlier.
Surface guidance systems track the patient's external surface topology in real time using optical cameras, providing a non-invasive surrogate for respiratory motion that can trigger gating or alert the team to unexpected motion. Increasingly, SGRT is replacing traditional tattoo-based alignment as the primary set-up and monitoring tool for thoracic radiotherapy.
Understanding the clinical problem - having measured it directly in patients, having worked through the physics of respiratory motion and phase binning, having contoured the volumes by hand - is what makes building AI solutions for it meaningful. The data in my thesis is not just academic history; it is the clinical grounding for the tools I am building now.