Technique selection and technical developments for 2D dual-energy subtraction angiography on an interventional C-arm

Abstract

Background: Dual-energy (DE) x-ray image acquisition has the potential to provide material-specific angiographic images in the interventional suite. This approach can be implemented with novel detector technologies, such as dual-layer and photon-counting detectors. Alternatively, DE imaging can be implemented on existing systems using fast kV-switching. Currently, there are no commercially available DE options for interventional platforms.

Purpose: This study reports on the development of a prototype fast kV-switching DE subtraction angiography system. In contrast to alternative approaches to DE imaging in the interventional suite, this prototype uses a clinically available interventional C-arm equipped with special x-ray tube control software. An automatic exposure control algorithm and technical features needed for such a system in the interventional setting are developed and validated in phantom studies.

Methods: Fast kV-switching was implemented on an interventional C-arm platform using software that enables frame-by-frame specification of x-ray tube techniques (e.g., tube voltage/kV, pulse width/ms, tube current/mA). A real-time image display was developed on a portable workstation to display DE subtraction images in real-time (nominal 15 frame/s). An empirical CNR-driven automatic exposure control (AEC) algorithm was created to guide DE tube technique selection (kV pair, ms pair, mA). The AEC model contained a look-up table which related DE tube technique parameters and air kerma to iodine CNR, which was measured in acrylic phantom models containing an iodine-equivalent reference object. For a given iodine CNR request, the AEC algorithm estimated patient thickness and then selected the DE tube technique expected to deliver the requested CNR at the minimum air kerma. The AEC algorithm was developed for DE imaging performed without and with the application of anti-correlated noise reduction (ACNR). Validation of the AEC model was performed by comparing the AEC-predicted iodine CNR values with directly measured values in a separate phantom study. Both dose efficiency (CNR²/kerma) and maximum achievable iodine CNR (within tube technique constraints) were quantified. Finally, improvements in DE iodine CNR were quantified using a novel variant to the ACNR approach, which used machine-learning image denoising (ACNR-ML).

Results: The prototype system provided a continuous display of DE subtraction images. For standard DE imaging, the AEC-predicted iodine CNR values agreed with directly measured values to within 3.5% ± 1.6% (mean ± standard deviation). When ACNR was applied, predicted iodine CNR agreed with measurement to within 2.1% ± 3.3%. AEC-generated DE techniques were typically (low/high energy): 63/125 kV, 10/3.2 ms, with varying mA values. When ACNR was applied, dose efficiency was increased by a factor of 9.37 ± 2.08 and maximum CNR was increased by a factor of 3.29 ± 0.21 relative to DE without denoising. Application of ACNR-ML yielded a greater increase in both the dose efficiency (16.11 ± 2.99) and maximum CNR (4.46 ± 0.31) compared to DE without denoising.

Conclusion: A prototype DE subtraction angiography system using fast kV-switching was implemented on a clinically available interventional C-arm platform without modification of system hardware. The technical features presented in this work include a real-time image display, noise-reduction strategies, and a CNR-driven AEC algorithm. This prototype system demonstrates the feasibility of 2D dual-energy imaging for image-guided interventions.

Keywords: angiography; dual‐energy; interventional imaging.

FIGURE 1

(a) Siemens Artis zee interventional C‐arm platform. (b) Portable workstation situated near the C‐arm system, containing the real‐time display (showing a DE image of a phantom) and a user‐interface. (c) User‐interface which contains certain features related to the DE image processing, including (1) the ability to view live images or load from file, (2) adjust the material‐canceling weighting factor, and (3) automatic image window/level adjustment.

FIGURE 2

Overview of the AEC model used to generate fast kV‐switching DE techniques. Both the iodine CNR request (determined prior to the study) and the estimated patient thickness (determined at the start of the study) are inputs to the AEC model. The AEC then finds all DE techniques which yield the CNR request for the given patient thickness using a look‐up table. Finally, the AEC model selects the DE technique with the minimum air kerma (if there are valid techniques), or the DE technique which yielded the highest CNR value (if there are no valid techniques).

FIGURE 3

X‐ray tube rating charts for single‐energy acquisitions, using the 1.0 mm focal spot size. The instantaneous limits show the maximum tube current as a function of tube voltage. The serial imaging limits define the maximum power as a function of duty cycle, scene time, and focus spot size for single‐energy imaging.

FIGURE 4

(a) Example phantom model used in creation of lookup table. The reference object is positioned in the center of the phantom within the air gap. Additional acrylic blocks placed on top of phantom facilitate determination of acrylic/tissue‐canceling DE weighting factors. (b) Example low‐energy image of phantom. Arrows point to the two regions used for material‐cancellation, including the reference object used for measurements of iodine CNR. (c) Example tissue/acrylic only DE image. (d) Example iodine/bone‐only DE image, which was used for measurements of iodine CNR.

FIGURE 5

Interpolation schemes used in AEC models. For all graphs, an origin point (0, 0) was specified for reference. (a) Tube current product (mAs) pair interpolation uses a 2D spline interpolation versus CNR measurements to create a surface of CNR values. (b) Interpolation between acrylic thickness values was accomplished by scaling surfaces of CNR measurements (fixed kV pair) according to an empirical relationship describing CNR versus phantom thickness, which was separately measured for a single mAs pair. The scaled CNR surfaces were then averaged to generate a surface of CNR measurements at the interpolated thickness value. (c) Interpolation across kV pairs was achieved by plotting CNR measurements as a function of kV pair (kV low, kV high) for fixed combinations of mAs pair and phantom thickness. 2D spline interpolation was then performed to generate a surface of CNR measurements as a function of DE kV pair.

FIGURE 6

(a) Anthropomorphic chest phantom. (b) Low energy image, showing two 2 mm ID plastic tubes containing iodinated contrast agent and saline. (c) Tissue/acrylic‐only DE image, which has been frame‐averaged (n = 16). The image signal is associated with the greater acrylic thickness in the spine and abdominal regions as well as the plastic tube walls. (d) Iodine/bone‐only image, which has been frame‐averaged (n = 16). The image signal is due to iodine in the tubes, the iodine reference object, and bones.

FIGURE 7

Validation of the AEC model, showing AEC‐predicted CNR (closed circles) versus achieved CNR (boxplots) in phantom models. Each subplot represents a unique CNR request specified to the AEC. Each boxplot represents the interquartile range (IQR) of the direct measurements (n = 20). Median values are shown with solid horizontal lines, and outliers are represented using plus signs. CNR measured using AEC‐generated DE techniques (blue boxplots) are compared with DE techniques generated using the ACNR algorithm (DE + ACNR; orange boxplots). The ACNR‐ML algorithm (DE + ACNR‐ML; yellow boxplots) was used to explore CNR increases relative to the ACNR algorithm. AEC‐predicted CNR values differ from requested values if the CNR was unachievable due to tube output constraints.

FIGURE 8

(Left) Dose efficiencies of DE tube techniques generated during the AEC validation study. Each boxplot contains dose efficiencies calculated across all three CNR requests (n = 60). (Right) Maximum achieved CNR, as a function of phantom thickness. Boxplots show the spread of direct measurements (n = 20). Dashed lines indicate smoothing splines of AEC predictions.

FIGURE 9

Iodine/bone‐only DE images taken of the chest phantom. Columns show processing technique; rows correspond to higher dose (top row), and lower dose (bottom row) acquisitions, where dose is reported as air kerma at the interventional reference point. ROIs show the reference object, ribs, and contrast‐filled tubes in an abdominal region with a 21.0 cm acrylic‐equivalent thickness.