Setup error in radiotherapy: on-line correction using electronic kilovoltage and megavoltage radiographs

We hypothesize that the difference in image quality between the traditional kilovoltage (kV) prescription radiographs and megavoltage (MV) treatment radiographs is a major factor hindering our ability to accurately measure, thus correct, setup error in radiation therapy. The objective of this work is to study the accuracy of on-line correction of setup errors achievable using either kV- or MV-localization (i.e., open-field) radiographs.

Using a gantry mounted kV and MV dual-beam imaging system, the accuracy of on-line measurement and correction of setup error using electronic kV- and MV-localization images was examined based on anthropomorphic phantom and patient imaging studies. For the phantom study, the user's ability to accurately detect known translational shifts was analyzed. The clinical study included 14 patients with disease in the head and neck, thoracic, and pelvic regions. For each patient, 4 orthogonal kV radiographs acquired during treatment simulation from the right lateral, anterior-to-posterior, left lateral, and posterior-to-anterior directions were employed as reference prescription images. Two-dimensional (2D) anatomic templates were defined on each of the 4 reference images. On each treatment day, after positioning the patient for treatment, 4 orthogonal electronic localization images were acquired with both kV and 6-MV photon beams. On alternate weeks, setup errors were determined from either the kV- or MV-localization images but not both. Setup error was determined by aligning each 2D template with the anatomic information on the corresponding localization image, ignoring rotational and nonrigid variations. For each set of 4 orthogonal images, the results from template alignments were averaged. Based on the results from the phantom study and a parallel study of the inter- and intraobserver template alignment variability, a threshold for minimum correction was set at 2 mm in any direction. Setup correction was applied by translating the treatment couch in the lateral, superior-to-inferior and vertical directions only. During treatment, kV open-field images were acquired for off-line treatment verification and analysis. Each patient study spanned 2-6 weeks. The 14 patient studies were completed with 8248 electronic images acquired and analyzed.

Results from the phantom studies showed that the users were able to detect the applied translational shift to better than 2 mm, and mostly to within 1 mm. The intraobserver variability of template alignment was on the order of 1 mm using a sample of either MV or kV patient images. The difference between using MV or kV images was significant for only a few cases. However in most cases, interobserver alignment variability was larger when using MV images than kV. For on-line setup correction, the study procedure added 10 min. to conventional treatment time. Setup variation measured with either kV- or MV-localization images was similar. The initial magnitude of setup error was appreciable, with a mean displacement of about 6.6 +/- 2.4 mm for the 14 patients. On-line correction using either kV- or MV-localization images improved setup accuracy. Over all study patients, setup errors occurred with standard deviations greater than 2 mm in any direction with a frequency of 48% before correction, and were reduced to 16% after correction. On average, kV image-based correction reduced radial setup variation to 2.6 +/- 1.6 mm compared to the 3.3 +/- 1.8 mm attained using MV images. The difference detected between the kV and MV data was not statistically significant when averaged over all patients. However, for on-line corrections in the neck and thoracic regions, using kV-localization images reduced setup error significantly more than using MV images.

In our anatomic template alignment study, interobserver variability was smaller using kV images than MV images. Intraobserver variability was smaller for alignments on kV images