Abstract
This study evaluated the agreement of fiducial marker localization between two modalities--an electronic portal imaging device (EPID) and cone-beam computed tomography (CBCT)--using a low-dose, half-rotation scanning protocol. Twenty-five prostate cancer patients with implanted fiducial markers were enrolled. Before each daily treatment, EPID and half-rotation CBCT images were acquired. Translational shifts were computed for each modality and two marker-matching algorithms, seed-chamfer and grey-value, were performed for each set of CBCT images. The localization offsets, and systematic and random errors from both modalities were computed. Localization performances for both modalities were compared using Bland-Altman limits of agreement (LoA) analysis, Deming regression analysis, and Cohen's kappa inter-rater analysis. The differences in the systematic and random errors between the modalities were within 0.2 mm in all directions. The LoA analysis revealed a 95% agreement limit of the modalities of 2 to 3.5 mm in any given translational direction. Deming regression analysis demonstrated that constant biases existed in the shifts computed by the modalities in the superior-inferior (SI) direction, but no significant proportional biases were identified in any direction. Cohen's kappa analysis showed good agreement between the modalities in prescribing translational corrections of the couch at 3 and 5 mm action levels. Images obtained from EPID and half-rotation CBCT showed acceptable agreement for registration of fiducial markers. The seed-chamfer algorithm for tracking of fiducial markers in CBCT datasets yielded better agreement than the grey-value matching algorithm with EPID-based registration.