Position detection accuracy of a novel linac-mounted intrafractional x-ray imaging system

Abstract

The authors have developed a system that monitors intrafractional target motion perpendicular to the treatment beam with the aid of radioopaque markers by means of separating kV image and megavoltage (MV) treatment field on a single flat-panel detector. They equipped a research Siemens Artiste linear accelerator (linac) with a 41 × 41 cm(2) a-Si flat-panel detector underneath the treatment head. The in-line geometry allows kV (imaging) and MV (treatment) beams to share closely aligned beam axes. The kV source, usually mounted directly across from the flat-panel imager, was retracted toward the gantry by 13 cm to intentionally misalign kV and MV beams, resulting in a geometric separation of MV treatment field and kV image on the detector. Two consecutive images acquired within 140 ms (the first with MV-only and the second with kV and MV signal) were subtracted to generate a kV-only image. The images were then analyzed "online" with an automated threshold-based marker detection algorithm. They employed a 3D and a 4D phantom equipped with either a single radioopaque marker or three Calypso beacons to mimic respiratory motion. Measured room positions were either cross-referenced with a phantom voltage signal (single marker) or the Calypso system. The accuracy of the back-projection (from detected marker positions into room coordinates) was verified by a simulation study. A phantom study has demonstrated that the imaging framework is capable of automatically detecting marker positions and sending this information to the tracking tool at an update rate of 7.14 Hz. The system latency is 86.9 ± 1.0 ms for single marker detection in the absence of MV radiation. In the presence of a circular MV field of 5 cm diameter, the latency is 87.1 ± 0.9 ms. The total RMS position detection accuracy is 0.20 mm (without MV radiation) and 0.23 mm (with MV). Based on the evaluated motion patterns and MV field size, the positional accuracy and system latency indicate that this system is suitable for real-time adaptive applications.