Skip to main content
Radiotherapy

Radiotherapy

Markerless monitoring tracks central airways during radiotherapy

04 Oct 2018
Phantom measurements
Experimental setup, showing the 3D phantom on a moving treatment couch. (Courtesy: C Hazelaar et al Radiother. Oncol. 10.1016/jradonc.2018.08.007 ©2018 Elsevier).

Patients receiving stereotactic body radiotherapy (SBRT) for central lung tumours face an increased risk of central airway toxicity, which in some cases may be potentially life-threatening. Markerless 3D monitoring of the proximal bronchial tree (PBT) could help manage this risk by tracking the position and motion of airways relative to the tumour, while offering a more reliable tracking option than reflective or implanted markers. Accurate real-time position monitoring could alert users to interrupt radiation treatment if the PBT position deviates beyond a pre-determined threshold.

Researchers at VU University Medical Center have investigated PBT motion monitoring using a markerless template matching and triangulation technique, based on kilovoltage (kV) projection images acquired during linac gantry rotation. Their study demonstrated that the technique is technically feasible for use during central lung SBRT (Radiother. Oncol. 10.1016/jradonc.2018.08.007).

Lead investigators Max Dahele and Wilko Verbakel from VUmc’s department of radiation oncology, together with doctoral student Colien Hazelaar and colleagues, conducted a feasibility study of markerless 3D position monitoring using kV projection images, which offer the capability for simultaneous visualization of tumour and airway position. They examined cone-beam CT (CBCT) projections of a 3D printed thorax phantom and clinical data, and employed prototype software for template generation, template matching and triangulation to determine bronchus motion in three dimensions.

Four-dimensional CT is used for treatment planning to determine a patient’s breathing motion and relative positions of organs-at-risk (OAR). However, these positions may change during the actual treatment. Internal and external markers can be used to monitor motion during treatment, but any marker is just a surrogate and may not capture the actual motion of the airway. This is not the case with real-time markerless position monitoring of the airways themselves.

Matching motion

The researchers created a 3D printed phantom consisting of soft tissue, bony structures, airways, lungs with blood vessels, and three lung tumours (Med. Phys. 45 92). They simulated irregular breathing motion by automatically moving the treatment couch, and acquired kV images for a full arc at 15 frames/second. Two structures, representing the left and right PBTs, were delineated on the planning CT scan. They also performed fluoroscopy during MV irradiation, acquiring images at 7 frames/second.

Airway motion

In addition, the researchers retrospectively analysed full-fan CBCT images (without irradiation, with 470-500 images per dataset) from 10 patients who had undergone free-breathing stereotactic or hypofractionated lung irradiation. All patients had tumours located within the field-of-view of a full-fan CBCT scan, with half of the tumours on the right side of the lung. The PBTs were similarly delineated.

The authors used prototype software (template-based tracking and the Sequential Stereo algorithm) to create 2D reference templates for every degree of gantry rotation. The templates included the delineated airways plus a 4 mm isotropic margin to include the airway walls. They selected the template associated with the gantry angle closest to the projection image. They matched the templates to the projection images for 2D PBT position using normalized cross-correlation; and used multiple registrations triangulated to determine the 3D position.

The right PBT had a better matching performance than the left PBT, which the authors attributed to the mediastinum being more obscured on the left side. In the phantom, the 2D right/left PBT position could be determined in 86.6%/75.1% of the CBCT datasets without MV irradiation; the 3D position (excluding the first 20°) was determined in 84.7%/72.7% of datasets.

In the patient dataset, there were no remarkable differences in matching performance between the left and right PBT, with 2D position determined in 89.8% of each dataset, 3D position in 76.4%, and 3D position (excluding the first 20°) in 85.1%. The authors suggested that performance could potentially be improved by analysing more data and creating more optimal triangulation parameter settings. They note that dynamic adjustment of kV and mA might offset poorer results of template matching when image quality was poor. They also advised that a different structure with distinguishable features could be used, such as branches of the PBT.

“We are currently using the same software for continuous spine position tracking during spine SBRT delivery,” Verbakel tells Physics World. “It is updated sub-second, and the accuracy is in the order of 0.3 mm. If the spine position during radiation delivery is more than 1 mm off, we interrupt the treatment for a new setup CBCT scan (Int. J. Radiat. Oncol. Biol. Phys. 101 1253).”

“We can also use the software for lung tumour position monitoring during breath-hold lung SBRT delivery,” he explains. “Since there can be large inter-breath-hold variation, we like to monitor if the tumour is indeed within the PTV. In case the tumour moves outside the PTV, we interrupt the treatment and do a new breath-hold, or a new setup CBCT scan (Phys. Med. Biol. 63 115005).”

Copyright © 2024 by IOP Publishing Ltd and individual contributors