2986 - Optimization of Acquisition and Reconstruction Parameters for Cranial SRS Localization Using a Next-Generation CBCT System on a C-Arm Linac

Purpose/Objective(s):

To determine optimal acquisition and reconstruction parameters for cranial SRS localization using a next-generation CBCT on a C-arm linac. The system features a larger 43×43 cm² detector, redesigned scatter grid, enhanced detector DQE, and advanced iterative recon. We evaluated the impact of rotation arc length (half [H] vs full [F]), recon algorithm (standard FDK [S] vs iterative [I]), and recon slice thickness (1 vs 2 mm) on image quality and 6D localization accuracy and precision. We hypothesized that full-rotation acquisition, iterative recon, and reduced slice thickness would minimize 3D localization error for bony alignment.

Materials/Methods:

Catphan image quality (noise, contrast-to-noise ratio [CNR], uniformity, spatial resolution) was measured for all parameter combinations using default head protocol kV-source settings. For localization assessment, an anthropomorphic head phantom was CT simulated. CBCTs for each parameter combination were acquired with known couch offsets (+1 mm translations, +1° rotations) applied in six degrees of freedom. Rigid registrations to the planning CT were performed, and localization error was defined as the difference between applied and registration-derived shifts. Primary endpoints were mean 3D translational and rotational localization error. Repeatability was assessed by calculating the standard deviation (SD) across three scans.

Results:

Full-rotation acquisition and iterative recon markedly improved image quality. Compared with H/S/2 (noise 2.01%, CNR 5.37, uniformity 96.7%), F/I/2 reduced noise to 0.77%, increased CNR to 13.87, and improved uniformity to 98.5%.

Mean 3D translational localization error ranged from 0.07 mm (F/I/1) to 0.20 mm (H/I/1). Holding slice thickness at 2 mm, F/I recon reduced translational error relative to H/I (0.12 mm vs 0.19 mm), with H/S demonstrating intermediate performance (0.15 mm). Full-rotation iterative recon at 1 mm achieved the lowest mean translational error (0.07 mm) and the smallest observed variability (SD 0.01 mm).

Reducing recon slice thickness from 2 mm to 1 mm improved translational accuracy for F/I scans (0.12 ± 0.06 mm vs 0.07 ± 0.01 mm) but increased noise (0.77% to 1.26%) and decreased CNR (13.87 to 8.60). In contrast, 1 mm recon did not improve accuracy for H/I.

Mean 3D rotational localization error ranged from 0.08° to 0.12° and did not vary meaningfully with acquisition or recon parameters.

Conclusion:

Systematic evaluation of acquisition and recon parameters demonstrated that a full-rotation with iterative recon yielded the lowest mean translational localization errors. Within this setting, decreasing recon slice thickness further improved translational accuracy and precision despite degraded image quality. Observed localization errors were well below recommended SRS delivery tolerances, demonstrating robust localization performance across all acquisition strategies with this next-generation CBCT imaging system.