To address this issue, PSI has equipped Gantry2 with a novel X-ray imaging system that acquires beam's eye view (BEV) images in the same direction as the treatment field. The aim is to use this BEV imaging system to monitor tumour motion during scanned beam delivery, and ultimately to enable real-time treatment adjustments. At the recent ESTRO 31 meeting in Barcelona, Spain, PSI's Ye Zhang described the latest developments in this project.
The BEV imaging system, she explained, uses an X-ray tube mounted on the back of the final 90° bending magnet, with the X-rays passing through a hole in the return yoke. The EPID detector panel is mounted under the patient bed on an extractable support. The source-to-detector distance is 444 cm and the source-to-isocentre distance is 395 cm, providing a field size at the isocentre of 20 x 25 cm.
"So how can we use this innovative imaging system to guide scanned proton therapy?" asked Zhang. Tumour motion during treatment – due to respiration, for example – is often addressed by expanding the clinical target volume to an internal target volume (ITV) that encompasses the motion envelope. The ultimate aim of image-guided treatment, explained Zhang, to reduce the size of this ITV.
The PSI team has developed algorithms that perform automatic recognition of multiple fiducial markers and segmentation of diaphragm images. Zhang noted that, in simulation studies, both diaphragm and marker motion could be observed and extracted, implying that this information could be used to guide gating or breath-hold during treatment.
To examine the system's potential for intra-fractional tumour motion tracking, the team generated time-resolved digitally reconstructed radiographs (DRRs) from simulated 4DCT datasets (static CT images with motion information from 4D MRI data) of liver cancer patients. These contained motion information over a relatively long time, including breath-hold and irregular breathing patterns. The DRRs, an approximation of the expected BEV images, were used to simulate a patient scenario, using the actual system geometry taken from calibration results.
In addition, from commissioning results of BEV imaging system, the accuracy of positional information obtained from the BEV image was within 0.5 mm for all possible gantry angles.
In the future, said Zhang, the system could potentially be used for "slow tracking", or motion management of the target's baseline drift, by performing BEV-guided breath-hold combined with rescanning. This process involves recording an initial surface image to gauge the respiratory signal, performing breath hold at the correct phases and acquiring the BEV image. This image is compared to DRR images calculated from the treatment plan and the positional variation is determined. If this variation is within the specified tolerance, the beam is switched on; if not, the above process is repeated.
"Our beam's-eye view imaging system is functional and our online motion extraction algorithms are working for both fiducial marker recognition and diaphragm segmentation," Zhang concluded. "This concept of BEV image guided "slow tracking" strategy is feasible, but still needs more work."