A mobile isocentric C-arm for intraoperative cone-beam CT: Technical assessment of dose and 3D imaging performance

Abstract

Purpose: To characterize the radiation dose and three-dimensional (3D) imaging performance of a recently developed mobile, isocentric C-arm equipped with a flat-panel detector (FPD) for intraoperative cone-beam computed tomography (CBCT) (Cios Spin 3D, Siemens Healthineers) and to identify potential improvements in 3D imaging protocols for pertinent imaging tasks.

Methods: The C-arm features a 30 × 30 cm² FPD and isocentric gantry with computer-controlled motorization of rotation (0-195°), angulation (±220°), and height (0-45 cm). Geometric calibration was assessed in terms of 9 degrees of freedom of the x-ray source and detector in CBCT scans, and the reproducibility of geometric calibration was evaluated. Standard and custom scan protocols were evaluated, with variation in the number of projections (100-400) and mAs per view (0.05-1.65 mAs). Image reconstruction was based on 3D filtered backprojection using "smooth," "normal," and "sharp" reconstruction filters as well as a custom, two-dimensional 2D isotropic filter. Imaging performance was evaluated in terms of uniformity, gray value correspondence with Hounsfield units (HU), contrast, noise (noise-power spectrum, NPS), spatial resolution (modulation transfer function, MTF), and noise-equivalent quanta (NEQ). Performance tradeoffs among protocols were visualized in anthropomorphic phantoms for various anatomical sites and imaging tasks.

Results: Geometric calibration showed a high degree of reproducibility despite ~19 mm gantry flex over a nominal semicircular orbit. The dose for a CBCT scan varied from ~0.8-4.7 mGy for head protocols to ~6-38 mGy for body protocols. The MTF was consistent with sub-mm spatial resolution, with f₁₀ (frequency at which MTF = 10%) equal to 0.64 mm^-1 , 1.0 mm^-1 , and 1.5 mm^-1 for smooth, standard, and sharp filters respectively. Implementation of a custom 2D isotropic filter improved CNR ~ 50-60% for both head and body protocols and provided more isotropic resolution and noise characteristics. The NPS and NEQ quantified the 3D noise performance and provided a guide to protocol selection, confirmed in images of anthropomorphic phantoms. Alternative scan protocols were identified according to body site and task - for example, lower-dose body protocols (<3 mGy) sufficient for visualization of bone structures.

Conclusion: The studies provided objective assessment of the dose and 3D imaging performance of a new C-arm, offering an important basis for clinical deployment and a benchmark for quality assurance. Modifications to standard 3D imaging protocols were identified that may improve performance or reduce radiation dose for pertinent imaging tasks.

Keywords: C-arm; cone-beam CT; image quality; image-guided surgery; radiation dose.

Fig. 1.

Experimental setup of the mobile C-arm for two-dimensional and three-dimensional imaging. (a) Dosimetry setup for adult body protocol (32 cm diameter phantom). (b)–(c) System geometry illustrating the world coordinate reference frame (w), virtual detector coordinate reference frame (i), and relevant geometric calibration parameters, as described in Cho et al. The detector rotation (ϕ_d, θ_d, η_d) describe the orientation of the detector relative to the virtual coordinate frame (i).

Fig. 2.

Geometric calibration and reproducibility for N = 10 orbits. Gantry (orbital) angles of 0° and 90° correspond to the x-ray beam directed along the x_w axis and y_w axis [Fig. 1(b)], respectively; for example, (for a patient laying supine on the OR table) a gantry angle of 0° corresponds to a lateral projection. Each orbit acquired 400 projections. (a) Variations in piercing point location on the detector. (b) Variations in source-detector distance over the course of the orbit. Deviations in (c) source position, (d) detector position, and (e) detector rotation over the course of a circular orbit. (f) Full width at half maximum of the point-spread function for various geometric calibration methods. All parameters prefixed with a “Δ” are plotted as the difference from a sinusoidal fit over the orbit. The shaded regions illustrate the variability (standard deviation) of the given parameter over the course of the orbit.

Fig. 3.

Uniformity and correspondence to Hounsfield units (HU). Slices of the uniform module in (a) axial and (b) sagittal planes (head protocol H with FBP₁ Normal filter). The yellow line in (b) corresponds to the axial slice shown in (a). (c) Noise map corresponding to (a). (d) AP and Lateral line profiles showing the magnitude of cupping artifact. (e) Uniformity (t_cup) for all clinical protocols in AP and LAT directions. (f) Correspondence of voxel value for various materials of known HU values. The dashed line represents the identity line.

Fig. 4.

Pre-sampling modulation transfer function (MTF) for each reconstruction filter. Results were equivalent for each of the clinical protocols in Table II (H protocol shown). Error bars represent the standard deviation among multiple measurements (N = 6). The MTFs measured for Normal and Normal two-dimensional filters show small differences near the tails due to slight differences in the roll-off between the Shepp-Logan and Hann filters.

Fig. 5.

Contrast-to-noise ratio for a simulated soft-tissue stimulus (−90 HU contrast) imaged at each clinical protocol and filter choice. (a) Head setup. (b) Body setup. Results for the manufacturer-specified FBP₁ algorithm are shown in color, and those for the FBP₂ algorithm (Normal filter, labeled “Normal 2D”) are shown in black.

Fig. 6.

Noise-power spectrum (NPS). (Top) Axial and (Bottom) longitudinal NPS for each clinical protocol and reconstruction filter for FBP₁. The NPS for FBP₂ (Normal filter) is shown in black, denoted “Normal 2D.” The NPS is shown for frequencies up to the Nyquist (1.6 mm⁻¹).

Fig. 7.

Axial noise-equivalent quanta for each various protocols and filters. Results for the FBP₁ algorithm various filters are in color, and for the FBP₂ algorithm (Normal 2D) are in black.

Fig. 8.

Cone-beam computed tomography images (sagittal) in the region of the temporal bone of an anthropomorphic head phantom. The visual image quality with respect to bone visualization reflects the modulation transfer function results in Fig. 4 and helped to guide technique chart definition — viz., head protocol H with Normal reconstruction filter for visualization of bone details.

Fig. 9.

Axial images of the anthropomorphic head phantom, illustrating the visibility of low-contrast stimuli for various scan protocols and reconstruction filters. A relatively low-contrast stimulus (−80 HU) is marked by the arrow, with contrast-to-noise ratio noted in the lower-left of each panel. The results help to guide development of future protocols that may support soft-tissue visualization in head imaging — for example, Custom Head 2 with Smooth or Normal two-dimensional filters.

Fig. 10.

Soft-tissue visibility in an anthropomorphic abdomen phantom. The higher x-ray tube output of the clinical body protocols provide improved soft-tissue visibility (e.g., the −95 HU contrast lesion marked by the arrow), with further improvement gained by the FBP₂ algorithm with isotropic two-dimensional filter.

Fig. 11.

Imaging performance for body scan protocols visualized in cone-beam computed tomography images of a pelvis phantom: coronal views of the region about the acetabulum (white arrow) and zoomed-in axial views (yellow inset) of a low-contrast insert (yellow arrow). The clinical body protocols (right three columns) deliver relatively high mAs (potentially suitable to soft-tissue visualization—for example, the H protocol with Smooth or Normal two-dimensional filters) — and motivated the investigation of custom body protocols (left three columns) for bone visualization at reduced radiation dose.