BackgroundAim of this study was to comparatively evaluate the accuracy of respiration–correlated (4D) and uncorrelated (3D) cone beam computed tomography (CBCT) in localizing lipiodolized hepatocellular carcinomas during stereotactic body radiotherapy (SBRT).Methods4D–CBCT scans of eighteen HCCs were acquired during free–breathing SBRT following trans–arterial chemo–embolization (TACE) with lipiodol. Approximately 1320 x–ray projections per 4D–CBCT were collected and phase–sorted into ten bins. A 4D registration workflow was followed to register the reconstructed time–weighted average CBCT with the planning mid–ventilation (MidV) CT by an initial bone registration of the vertebrae and then tissue registration of the lipiodol. For comparison, projections of each 4D–CBCT were combined to synthesize 3D–CBCT without phase–sorting. Using the lipiodolized tumor, uncertainties of the treatment setup estimated from the absolute and relative lipiodol position to bone were analyzed separately for 4D– and 3D–CBCT.ResultsQualitatively, 3D–CBCT showed better lipiodol contrast than 4D–CBCT primarily because of a tenfold increase of projections used for reconstruction. Motion artifact was observed to subside in 4D–CBCT compared to 3D–CBCT. Group mean, systematic and random errors estimated from 4D– and 3D–CBCT agreed to within 1 mm in the cranio–caudal (CC) and 0.5 mm in the anterior–posterior (AP) and left–right (LR) directions. Systematic and random errors are largest in the CC direction, amounting to 4.7 mm and 3.7 mm from 3D–CBCT and 5.6 mm and 3.8 mm from 4D–CBCT, respectively. Safety margin calculated from 3D–CBCT and 4D–CBCT differed by 2.1, 0.1 and 0.0 mm in the CC, AP, and LR directions.Conclusions3D–CBCT is an adequate alternative to 4D–CBCT when lipoid is used for localizing HCC during free–breathing SBRT. Similar margins are anticipated with 3D– and 4D–CBCT.

Objective: To evaluate the effectiveness of a patient-specific immobilization and positioning device in prostate radiotherapy. Methods: Eighty patients were immobilized and positioned by a patientspecific device. Prostate translations and rotations were estimated from daily cone beam computed tomography scans using a contour-based approach assisted by auto-registration and quantified by the group mean GM, systematic Σ and random σ' errors. Dosimetric impacts of residual prostate rotations where the translation errors were corrected were evaluated by robustness plan analysis. Results: Using the patient-specific immobilization alone without online image-guidance, the GM, Σ and σ' of the prostate translations were 0.8, 1.7, and 1.5 mm (left-right; LR), 0.8, 2.1, and 1.9 mm (superior-inferior; SI), and 0.5, 1.7 and 1.5 mm (anterior-posterior; AP), while for the prostate rotations they were 0.0˚, 0.6˚, and 0.7˚ (pitch), 0.2˚, 0.5˚, and 0.6˚ (roll), and 0.2˚, 0.5˚, and 0.6˚ (yaw). The resulting van Herk's margin was 5.8 (LR), 7.3 (SI) and 5.8 (AP) mm. With adaptive online imageguidance based on estimates from the first 5 fractions, Σ were reduced by 0.7-1.2 mm for the prostate translations, resulting in a margin reduction by 2-3.5 mm. Changes of Σ and σ' in the prostate rotations were insignificant regardless of translation corrections. Dosimetric impacts of residual rotation errors were negligible if a 2 mm margin was applied. Conclusions: Our patientspecific immobilization system can effectively limit the prostate translations and rotations, which is important without 6D treatment couches or using ultrasound image-guidance without rotational corrections.