Interleaved x-ray imaging: A method for simultaneous acquisition of quantitative and diagnostic digital subtraction angiography

Abstract

Background:

Flow altering angiographic procedures suffer from ill-defined, qualitative endpoints. Quantitative digital subtraction angiography (qDSA) is an emerging technology that aims to address this issue by providing intra-procedural blood velocity measurements from time-resolved, 2D angiograms. To date, qDSA has used 30 frame/s DSA imaging, which is associated with high radiation dose rate compared to clinical diagnostic DSA (up to 4 frame/s).

Purpose:

The purpose of this study is to demonstrate an interleaved x-ray imaging method which decreases the radiation dose rate associated with high frame rate qDSA while simultaneously providing low frame rate diagnostic DSA images, enabling the acquisition of both datasets in a single image sequence with a single injection of contrast agent.

Methods:

Interleaved x-ray imaging combines low radiation dose image frames acquired at a high rate with high radiation dose image frames acquired at a low rate. The feasibility of this approach was evaluated on an x-ray system equipped with research prototype software for x-ray tube control. qDSA blood velocity quantification was evaluated in a flow phantom study for two lower dose interleaving protocols (LD1: $3.7\hspace{0.17em}\pm \hspace{0.17em}0.02 \hspace{0.17em}mGy/s$ and LD2: $1.7\hspace{0.17em}\pm \hspace{0.17em}0.04\hspace{0.17em}mGy/s$) and one conventional (full dose) protocol ($11.4\hspace{0.17em}\pm \hspace{0.17em}0.04\hspace{0.17em}mGy/s$). Dose was measured at the interventional reference point. Fluid velocities ranging from 24 to 45 cm/s were investigated. Gold standard velocities were measured using an ultrasound flow probe. Linear regression and Bland-Altman analysis were used to compare ultrasound and qDSA.

Results:

The LD1 and LD2 interleaved protocols resulted in dose rate reductions of −67.7% and −85.5%, compared to the full dose qDSA scan. For the full dose protocol, the Bland-Altman limits of agreement (LOA) between qDSA and ultrasound velocities were [0.7, 6.7] cm/s with a mean difference of 3.7 cm/s. The LD1 interleaved protocol results were similar (LOA: [0.3, 6.9] cm/s, bias: 3.6 cm/s). The LD2 interleaved protocol resulted in slightly larger LOA: [−2.5, 5.5] cm/s with a decrease in the bias: 1.5 cm/s. Linear regression analysis showed a strong correlation between ultrasound and qDSA derived velocities using the LD1 protocol, with a $R^2$ of $0.96$, a slope of $1.05$ and an offset of $1.9$ cm/s. Similar values were also found for the LD2 protocol, with a $R^2$ of $0.93$, a slope of $0.98$ and an offset of $2.0$ cm/s.

Conclusions:

The interleaved method enables simultaneous acquisition of low-dose high-rate images for intra-procedural blood velocity quantification (qDSA) and high-dose low-rate images for vessel morphology evaluation (diagnostic DSA).

1. Introduction

Interventional procedures that aim to alter arterial blood flow (e.g. angioplasty, embolization, stent placement) employ x-ray angiography to evaluate treatment progress via changes in vessel morphology or visual patterns of blood flow. Often, these interventions employ qualitative angiographic endpoints with high inter-observer variability.¹ In the case of transarterial embolization, this variability is problematic since insufficient embolization will result in failure to achieve tumor necrosis, whereas over-embolization can induce angiogenesis and increase procedural toxicity and mortality.^1–3 Using quantitative endpoints based on the change of blood flow or velocity could help standardize these procedures and achieve more consistent procedure endpoints.

Approaches to providing quantitative information in flow altering procedures include 4D transcatheter intra-arterial perfusion MRI (TRIP-MRI), 4D x-ray imaging-based methods, doppler ultrasound, and 2D x-ray imaging-based methods. These techniques employ various quantitative metrics such as perfusion, time of arrival, time to peak, blood flow and blood velocity. Gaba et al. used TRIP-MRI to measure quantitative changes pre- and post transarterial chemoembolization (TACE) and found that TRIP-MRI successfully measured semi-quantitative changes in hepatocellular carcinoma perfusion during MR monitored TACE.⁴ 4D methods for an interventional x-ray system have been developed based on the concept of combining information from 2D time-resolved imaging and a separate 3D scan,^5–8 or, in the case of 4D-DSA,^9–17 by applying an alternative reconstruction scheme to the projection data from a 3D-DSA style acquisition. 4D-DSA has been shown to correlate with blood velocity in a hepatic swine model.^11,12 Additionally, doppler ultrasound can be used to quantify hepatic blood velocity¹⁸ during hepatic interventions. However, ultrasound has known limitations such as imaging depth, obstructions (e.g. bone or bowel gas), and angle of insonation effects.^19–21 A review article by Shpilfoygel et al.²² provides a comprehensive review of 2D x-ray imaging methods that use contrast dynamics to measure blood flow or velocity, or related metrics such as time-of-arrival, time-to-peak, and area under the curve. Recently, quantitative digital subtraction angiography (qDSA) has been proposed for intraprocedural quantification of blood velocity from 2D time resolved angiograms.^23–25 This approach has been correlated with measured mean blood velocity in swine liver, spleen and kidney models, and it employs commonly available x-ray equipment.

2D DSA imaging is routinely performed during hepatic interventions to perform visual assessments of the vascular structure and contrast dynamics. While conventional DSA in the body is performed at up to 4 frame/s (fps),^26,27 previous studies of qDSA relied on 30 fps DSA imaging due to the need for high temporal sampling of contrast dynamics.^23,24 A reduction in radiation dose rate is needed for the clinical translation of qDSA. Dose rate can be reduced either by lowering the frame rate or reducing the target detector dose per image frame, or through a combination of these techniques. Unfortunately, using conventional DSA frame rates results in insufficient temporal sampling for accurate qDSA. Low dose per frame, high frame rate (e.g. fluoroscopic quality) imaging can support qDSA needs. However, qDSA performed at substantially reduced dose per frame may not deliver the image quality needed for standard-of-care diagnostic assessment of vessel morphology. Given this limitation, qDSA at a low dose per frame would require an additional image acquisition and an additional contrast injection. The additional workflow step is undesirable, and the increased iodine dose is problematic for patients with reduced kidney function who are at risk for contrast-induced nephropathy.²⁸ Ideally, qDSA should provide the added value of quantitative blood velocity information without substantially increasing radiation dose, workflow steps, or contrast dose relative to a standard intervention. This is especially important if qDSA is to be used multiple times throughout a procedure to measure treatment progress.

The purpose of this work is to present a clinically feasible technique for simultaneous acquisition of contrast dynamics and anatomical information through the interleaving of high dose (clinical DSA equivalent) and low dose x-ray images in a single sequence, with a single contrast injection. This technique would provide both high rate images for intra-procedural quantification of blood velocity (qDSA) and high dose per frame images for assessment of vessel morphology (diagnostic angiography). The goal is to aid the clinical translation of qDSA by providing a seamless, non-disruptive integration into the current clinical workflow.

2. Materials and Methods

2.1. Interleaved Imaging

X-ray imaging was performed with a floor-mounted C-arm of an interventional x-ray angiography system (Artis zee, Siemens, Forchheim, Germany). The system was equipped with manufacturer-provided research software that allows frame-by-frame specification of x-ray tube voltage (kV), tube current (mA) and pulse width (ms). Interleaving of high and low radiation dose pulses can be achieved by modulating mAs and/or kV. Figure 1 gives a graphical representation of interleaving via mAs modulation, where the overall pulse rate is 25 pulse/s, every 10^th pulse is a high dose pulse yielding 2.5 fps diagnostic angiography, and the intervening 9 pulses are low dose pulses acquired for quantitative DSA. In this scheme, the low dose pulses are generated at 25 pulse/s, but accounting for the gaps created by the high dose pulses, the time-averaged frame rate for the low dose pulses is 22.5 fps. Currently, the prototype software performs frame-by-frame modulation only in a rotational acquisition mode. qDSA in its most basic form is meant to be a 2D image acquisition mode with fixed view angle. Therefore, phantom studies were performed in a pseudo-2D imaging setup, in which a circularly symmetric vessel phantom was positioned with its central axis coincident with the C-arm rotation axis to yield consistent 2D projections for any rotation angle.

Figure 1:

Example of an interleaved image acquisition. X-ray pulses are produced at 25 pulse/s. Every 10^th pulse is at high dose and the intervening 9 pulses are at low dose.

2.2. Flow Phantom

A flow phantom study was conducted to test the hypothesis that an interleaved technique preserves the ability of qDSA to quantify velocity (Figure 2). The vessel phantom consisted of a gelatin-filled acrylic cylinder (inner and outer diameter of 11.9 and 12.1 cm) with a 6.35 mm inner diameter tube along its central axis. To create the pseudo-2D imaging setup, a precision adjustment stage was used to align the central axis of the acrylic cylinder with the central axis of the C-arm. To provide realistic flow, the phantom was connected in line with a pulsatile displacement pump (BDC Laboratories, Wheat Ridge, CO) programmed to simulate a cardiac waveform at 60 beats per minute. Blood mimicking fluid consisting of 60:40 water:glycerol mix by volume was used for realistic fluid viscosity.²⁹ Contrast injections were performed using a Nemoto Press Duo power injector (Nemoto Kyorindo, Bunkyo-ku, Tokyo) and a 5 Fr flush-tipped catheter with the tip positioned 8.0 cm proximal to the tube entrance to the phantom. All images were acquired in a rotational acquisition mode consisting of a $-{45}^o$ to ${45}^o$ rotation in ${0.3}^o$/frame increments.

Figure 2:

Flow phantom experimental setup. A pulsatile flow pump was used for physiologically realistic cardiac waveforms. For pseudo-2D imaging, a fine stage adjustment was used to align the central axis of the cylindrical shell with the central axis of the C-arm rotation. An ultrasound flow probe was used to acquire gold standard flow measurements.

2.3. Velocity Quantification

The algorithm used to quantify blood velocity from 2D DSA images is detailed in the work by Wagner et al.³⁰ Briefly, blood velocity is computed by tracking oscillations in iodine contrast as they propagate along a vessel centerline. These oscillations are created naturally when contrast agent, injected at a constant rate, mixes with the time-varying blood flow driven by the cardiac cycle. A time-attenuation curve (TAC) capturing image signal vs. time is measured for each point along the vessel centerline. In practice, TAC signals are averaged across a 1.0 mm wide profile perpendicular to the vessel centerline to reduce noise. The time shift $t$ between the TACs measured for proximal and distal points of the same vessel segment represents the time required for the blood to flow between the two points. In theory, the average blood velocity $v$ between the two points can be determined as $v\hspace{0.17em}=\hspace{0.17em}d/t$, where $d$ is the pathlength between the proximal and distal point. In practice, the apparent pathlength along a vessel centerline may be measured from a 2D x-ray projection image, resulting in an apparent blood velocity, or a true blood velocity may be determined from a true vessel centerline length. A true vessel centerline length may be acquired with 3D imaging, or a catheter with markers of known spacing placed within the vessel of interest.²⁴

If TAC measurements are made for each point along a vessel centerline, the results can be assembled into a 2D time attenuation map (TAM) with time running along one axis and vessel centerline position along the other axis. When the TAM is viewed as a single image, the propagation of contrast along the vessel centerline appears as a sheared pattern. An example of the workflow used to measure TACs and a TAM is shown in Figure 3. The sheared pattern is visible in the TAM.

Figure 3:

Left) Example of a contrast-enhanced time-resolved 2D x-ray image of the phantom with corresponding measurement points and centerline. Middle) Individual time attenuation curves (TACs) measured from the time-resolved x-ray images. Right) Time attenuation map (TAM) constructed from TACs measured at every point along the tube centerline. The dark sheared lines in the TAM represent contrast traveling along the centerline over time.

The average velocity $v$ depends on the distance and time for the transit of the contrast pulse along the analyzed vessel segment, and therefore the slope $m$ of the sheared pattern:

$$\nu \hspace{0.17em}=\hspace{0.17em}\frac{d}{t}\hspace{0.17em}=\hspace{0.17em}m.$$ (Eq. 1)

Wagner et al. showed that this slope and velocity can be determined from the position of the fundamental frequency peak (corresponding to the oscillating contrast signal) in the 2D spatiotemporal Fourier transform of the TAM.³⁰ If the fundamental frequency peak appears at some temporal frequency ${\xi }_{\omega }\hspace{0.17em}=\hspace{0.17em}\omega$ then the spatial frequency of the peak is at ${\psi }_{\omega }\hspace{0.17em}=\hspace{0.17em}-\omega /m$, and the velocity is given by,

$$v\hspace{0.17em}=\hspace{0.17em}-\frac{{\xi }_{\omega }}{{\psi }_{\omega }}.$$ (Eq. 2)

2.4. Experimental Design

2.4.1. Interleaved Protocols

Two lower dose interleaved imaging protocols (LD1, LD2) were compared to a reference imaging protocol (REF) where full dose pulses are used at a high frame rate. The REF protocol used a tube technique of $74 \hspace{0.17em}\mathrm{kV,}\hspace{0.17em}2.59\hspace{0.17em}\mathrm{mAs}/f\mathrm{rame}$ for all frames and 25 fps (and 25 pulse/s) acquisition. The reference tube technique was determined by imaging the phantom with a clinical abdominal protocol with the system’s automatic exposure control active and the detector dose programmed to 3.6 μGy/frame. The interleaved protocols used the timing scheme shown in Figure 1, with 9 low dose pulses after each full dose pulse, and the same pulse rate as the REF protocol. The LD1 protocol used $74\hspace{0.17em}\mathrm{kV, }0.66\hspace{0.17em}\mathrm{mAs/}frame$ for the low dose pulses and the LD2 protocol used $74\hspace{0.17em}\mathrm{kV, }0.\mathrm{10}\hspace{0.17em}\mathrm{mAs/}frame$ for low dose pulses. Both used full dose pulses identical to the REF protocol. The LD1 protocol represented the maximum dose rate reduction that could be achieved on the current prototype using pulse width (ms) modulation alone. The LD2 protocol used modulation of mAs/frame for greater dose reduction.

For an interleaving protocol that relies on ms or mAs modulation, the expected mean dose rate reduction (DRR) relative to a reference acquisition performed with all full dose pulses is

$$DRR\left[\%\right]\hspace{0.17em}=\hspace{0.17em}100\%\times \left(\frac{{\mathrm{mAs}}_L{f}_L\hspace{0.17em}+\hspace{0.17em}{\mathrm{mAs}}_F{f}_F}{{\mathrm{mAs}}_F}-1\right),$$ (Eq. 3)

where ${\mathrm{mAs}}_F$ and ${\mathrm{mAs}}_L$ are the mAs for full dose and low dose pulses, respectively, and $f_F$ and $f_L$ are the fractions of pulses in the interleaved protocol performed at full- and low-dose, respectively. The expected DRRs for the LD1 and LD2 protocols were −67.1% and −86.5%. Since the LD2 protocol operates at 13.5% of the dose rate of a 25 fps acquisition with full dose pulses, it has a dose rate equivalent to 0.135 × 25 fps = 3.4 fps conventional DSA with full dose pulses, which is within the range of frame rates employed for clinical body DSA.^26,27

2.4.2. Flow Study

For each protocol, image sequences were acquired at five mean native fluid velocities (19, 24, 29, 35 and 41 cm/s prior to contrast injection) with eight repeat acquisitions per flow rate. These mean velocities were chosen to be at and slightly below the typical range for hepatic arterial velocity in humans (common hepatic: $51\pm 3$ cm/s, left hepatic: $28\pm 2$ cm/s, right hepatic: $35\pm 4$ cm/s),³¹ to simulate the effects of embolization which causes a decrease in blood flow. Iodine (300 mgI/mL iohexol) contrast injections were performed through the catheter at 2.0 mL/s for 7.0 s. Gold standard flow measurements were acquired during each image sequence using a Doppler ultrasound (US) flow probe (6PXL Transonic, Ithaca, NY) attached 40 cm distal to the catheter tip. To ensure accurate gold standard flow measurements, the US flow probe was calibrated using a five-point calibration performed via the timed volume method on the day of the experiment. For comparison with qDSA derived velocities, US measured flows were converted to velocity by dividing by the nominal cross-sectional area of the tube.

Image analysis was performed on DICOM images using MATLAB (MATLAB R2020a v.9.8, MathWorks, Natick, MA). Prior to qDSA analysis, images were normalized to correct for the large changes in brightness in neighboring full- and low-dose frames. Normalization was achieved by dividing all image values by the average pixel intensity of a background region of interest next to the central region of the phantom. qDSA analysis was performed on all image sequences to determine velocity. Centerline points for qDSA analysis were measured using a pixel pitch of 0.308 mm. Apparent velocity measurements were corrected to true velocity measurements using a magnification factor of 1.6. To approximate a realistic vessel length, qDSA measurements were performed in a 3.0 cm long segment of the tube 15 cm distal to the catheter tip. TACs were measured using image frames acquired 1.5 to 8.5 seconds after the start of the contrast injection to allow for stabilization of contrast dynamics.

2.4.3. Radiation Dose Measurement

Radiation dose measurements were made to ensure correct radiation dose modulation using the interleaving prototype. Dose measurements were made with a RadCal AccuGold meter with an $AGMS-D+$ sensor with a $102.4 \mu s$ sample period. The RadCal was attached to the x-ray detector to enable radiation dose measurements from a rotational acquisition mode, with the table and phantom removed from the beam path. To avoid saturation of the image receptor, $~1$ mm of lead was placed between the RadCal meter and the flat panel image receptor. An inverse square law correction was applied to scale the dose measurements to the interventional reference point located 15 cm toward the source from the isocenter. The distance from the x-ray source to the RadCal detector was 115 cm and the source to isocenter distance was 75.0 cm. For each of the three protocols, repeat radiation dose measurements were made on five image sequences. Dose rate calculations used the data from 1.5 to 8.5 seconds after the start of contrast injection.

Mean tube parameters (tube current, pulse width, tube voltage) for the full- and low-dose pulses for each protocol were also measured across repeat image sequences. The pulse width and tube voltage were directly measured using the RadCal AccuGold meter, while the tube current was determined using the system measured tube current.

2.5. Evaluation

2.5.1. Interleaved Imaging

Dose per pulse was determined by integrating the RadCal measured dose rate waveform for each pulse. Dose rate (mGy/s) was then computed as the measured dose per pulse multiplied by the pulse rate measured from the RadCal meter. The measured dose rate reduction (DRR) was calculated by

$$DRR\left[\%\right]\hspace{0.17em}=\hspace{0.17em}100\%\times \left(\frac{({\overline{\text{D}}}_L{f}_L\hspace{0.17em}+\hspace{0.17em}{\overline{\text{D}}}_F{f}_F)P_{int}}{{\overline{\text{D}}}_{ref}{P}_{ref}}-1\right),$$ (Eq. 4)

where ${\overline{D}}_L$ and ${\overline{D}}_F$ are the mean radiation dose/pulse for low- and full-dose pulses in the interleaved protocol, $f_L$ and $f_F$ are the fractions of low- and full-dose pulses in the interleaved protocol, ${\overline{D}}_{ref}$ is the mean radiation dose/pulse in the reference protocol, and $P_{int}$ and $P_{ref}$ are the measured pulse rates (pulse/s) in the interleaved and reference protocols.

The consistency in radiation dose delivery for interleaved imaging was evaluated using two metrics. The first metric examined inter-scan variability. To this end, a percent coefficient of variation ($\%CV$) was determined for each frame in an interleaved protocol using five repeat sequences. The mean and standard deviation were then computed from the $\%CV$ results for all frames. The second metric examined intra-scan variability. The %CV were computed separately for the full- and low-dose pulses within a single sequence, and then mean and standard deviation were computed over all sequences. The same analysis of inter-scan and intra-scan variability was repeated for the pulse-by-pulse kV measurements, to demonstrate kV consistency.

The full dose pulses of an interleaved protocol should have a dose level and kV matching the pulses in the reference full dose protocol. To evaluate this, a two one sided t-test (TOST) procedure for equivalence was used to test if the measured radiation dose in the full dose image frames was within $\pm 1\%$ of the average radiation dose per frame in the reference image sequences. The same test was then repeated for the tube voltage.

2.5.2. qDSA Velocity Measurement

To evaluate the agreement between US and qDSA derived velocity measurements, linear regression and Bland-Altman analyses were performed. Both analyses were performed for all three protocols using US as the gold standard. The linearity between US and qDSA velocity measurements was evaluated via the coefficient of determination ($R^2$) and the 95% confidence intervals (CI) for the slope and offset were computed. Bland-Altman analysis was used to investigate the mean difference and limits of agreement (LOA) between US and qDSA derived velocities.

3. Results

3.1. Interleaved Imaging

All results are reported as mean $\pm$ standard deviation and all radiation dose values are cited at the interventional reference point, unless otherwise stated. The measured radiation dose per frame at the interventional reference point for each protocol is shown in Figure 4. For the LD1 and LD2 protocols, the average radiation dose per pulse in the full dose pulses (LD1: $0.44\hspace{0.17em}\pm \hspace{0.17em}0.002 \hspace{0.17em}mGy/pulse$, LD2: $0.45\hspace{0.17em}\pm \hspace{0.17em}0.01\hspace{0.17em}mGy/pulse$) was within $1\%$ of the average radiation dose per pulse in the REF protocol ($0.45\hspace{0.17em}\pm \hspace{0.17em}0.002\hspace{0.17em}mGy/pulse$, $p\hspace{0.17em}<\hspace{0.17em}0.001$), confirming that the average desired dose level was consistently achieved in the full dose pulses.

Figure 4:

Radiation dose at the interventional reference point per frame for a reference (REF), and two lower dose interleaved imaging protocols (LD1 and LD2).

Table 1 shows mean radiation dose rates (mGy/s) for all three protocols. Mean dose rate reduction (DRR) for the LD1 and LD2 protocols were −67.7% and −85.5%, respectively, relative to the REF protocol. For comparison, the predicted DRRs were −67.1% and −86.5% based on the programmed mAs/pulse patterns in the interleaved protocols.

Table 1:

Average dose rates [mGy/s] for the REF, LD1 and LD2 qDSA protocols. Percent dose rate reduction (DRR) relative to the reference (REF). Results reported as mean $\pm$ standard deviation.

qDSA Protocol	Full Dose Frames [mGy/s]	Low Dose Frames [mGy/s]	Total Dose Rate [mGy/s]	Measured $\mathrm{DRR }\left[\%\right]$	Predicted $\mathrm{DRR }\left[\%\right]$
REF	$11.4\hspace{0.17em}\pm \hspace{0.17em}0.04$	-	$11.4\hspace{0.17em}\pm \hspace{0.17em}0.04$	-	-
LD1	$1.1\hspace{0.17em}\pm \hspace{0.17em}0.01$	$2.6\hspace{0.17em}\pm \hspace{0.17em}0.02$	$3.7\hspace{0.17em}\pm \hspace{0.17em}0.02$	−67.7	−67.1
LD2	$1.1\hspace{0.17em}\pm \hspace{0.17em}0.02$	$0.5\hspace{0.17em}\pm \hspace{0.17em}0.04$	$1.7\hspace{0.17em}\pm \hspace{0.17em}0.04$	−85.5	−86.5

The inter-scan %CV of radiation dose for the REF, LD1 and LD2 (n = 179) protocols were $0.3\hspace{0.17em}\pm \hspace{0.17em}0.1\%$, $0.6\hspace{0.17em}\pm \hspace{0.17em}0.2\%$ and $2.5\hspace{0.17em}\pm \hspace{0.17em}0.9\%$, respectively, demonstrating consistent radiation dose delivery for a given frame across different acquisitions. The intra-scan %CVs of the radiation dose for the LD1 protocol (n = 5) were $0.3\hspace{0.17em}\pm \hspace{0.17em}0.04\%$ and $0.7\hspace{0.17em}\pm \hspace{0.17em}0.03\%$ for the full- and low-dose pulses, respectively, and $1.8\hspace{0.17em}\pm \hspace{0.17em}0.09\%$ and $7.2\hspace{0.17em}\pm \hspace{0.17em}0.2\%$ for the LD2 protocol. These results demonstrate an increase in the average %CV value from 2.5% to 7.2% for the inter- and intra-scan variability of the LD2 protocol. The intra-scan %CV of the radiation dose for the REF protocol (n = 5) was $0.3\hspace{0.17em}\pm \hspace{0.17em}0.04\%$. The measured pulse/s was 25.4 Hz for all three protocols.

Table 2 shows measured tube parameters for all three protocols. For the LD1 and LD2 protocols, the average of the tube voltage in the full dose pulses (LD1: $74.6\hspace{0.17em}\pm \hspace{0.17em}0.1\hspace{0.17em}kV$, LD2: $74.8\hspace{0.17em}\pm \hspace{0.17em}0.1\hspace{0.17em}kV$) was within $1\%$ of the average of the tube voltage values in the REF protocol ($74.4\hspace{0.17em}\pm \hspace{0.17em}0.1\hspace{0.17em}kV$, $p\hspace{0.17em}<\hspace{0.17em}0.001$). The measured average kV was closely aligned with the requested $74.0\hspace{0.17em}kV$ for all protocols; however, the low dose frames in the LD2 protocol had a small decrease in the average kV with an increase in standard deviation. The inter-scan %CV in kV for a given image frame (n = 179) measured across acquisitions for the REF, LD1 and LD2 protocols were $0.08\hspace{0.17em}\pm \hspace{0.17em}0.03\%$, $0.2\hspace{0.17em}\pm \hspace{0.17em}0.1\%$ and $1.3\hspace{0.17em}\pm \hspace{0.17em}0.6\%$, respectively, demonstrating consistent kV for a given frame across different acquisitions. The intra-scan %CVs of the measured kV for the LD1 protocol (n = 5) were $0.13\hspace{0.17em}\pm \hspace{0.17em}0.03\%$ and $0.29\hspace{0.17em}\pm \hspace{0.17em}0.01\%$ for the full- and low-dose pulses, respectively, and $0.09\hspace{0.17em}\pm \hspace{0.17em}0.01\%$ and $2.8\hspace{0.17em}\pm \hspace{0.17em} 0.05\%$ for the LD2 protocol. The intra-scan %CV of the kV for the REF protocol (n = 5) was $0.09\hspace{0.17em}\pm \hspace{0.17em}0.01\%$.

Table 2:

Measured tube parameters (mean$\pm$standard deviation) per frame for the reference (REF) and two interleaved protocols LD1 and LD2.

qDSA Protocol	Full Dose Frames			Low Dose Frames
	kV	ms	mA	kV	ms	mA
REF	$74.4\hspace{0.17em}\pm \hspace{0.17em}0.1$	$12.4\hspace{0.17em}\pm \hspace{0.17em}0.05$	$251.3\hspace{0.17em}\pm \hspace{0.17em}0.1$	-	-	-
LD1	$74.6\hspace{0.17em}\pm \hspace{0.17em}0.1$	$12.4\hspace{0.17em}\pm \hspace{0.17em}0.05$	$251.2\hspace{0.17em}\pm \hspace{0.17em}0.1$	$74.3\hspace{0.17em}\pm \hspace{0.17em}0.2$	$3.1\hspace{0.17em}\pm \hspace{0.17em}0.05$	$257.8\hspace{0.17em}\pm \hspace{0.17em}0.5$
LD2	$74.8\hspace{0.17em}\pm \hspace{0.17em}0.1$	$9.8\hspace{0.17em}\pm \hspace{0.17em}0.04$	$323.6\hspace{0.17em}\pm \hspace{0.17em}6.6$	$72.7\hspace{0.17em}\pm \hspace{0.17em}2.0$	$0.5\hspace{0.17em}\pm \hspace{0.17em}0.06$	$168.4\hspace{0.17em}\pm \hspace{0.17em}7.5$

Figure 5 shows the measured tube parameters per frame for the REF and LD2 protocols. The pulse width and tube current during the LD2 protocol can be seen to quickly increase during the full dose pulses. As the tube current slowly decreases between full dose pulses, the pulse width gradually increases to maintain a consistent mAs during low dose pulses, as demonstrated in the mAs plot.

Figure 5:

Measured tube parameters for the REF and LD2 protocols as a function of frames used in the qDSA analysis.

3.2. qDSA Velocity Measurements

Figure 6 shows a comparison of a full dose image acquired during the LD2 protocol and a neighboring frame acquired at low dose. As expected, the relative noise level is increased in the low dose image.

Figure 6:

A) Image frame from a full dose (no interleaving) acquisition. B) Full dose image acquired during the LD2 interleaved protocol. C) Neighboring frame acquired at lower dose.

Figure 7 shows individual TACs measured for a point on the tube of the vessel phantom. The contrast oscillations have a period of 1 second, which corresponds to a single cycle of the 60 bpm simulated heart rate. The noise level in the TACs can also be seen to increase as the radiation dose decreases.

Figure 7:

Time attenuation curves (TACs) shown for the reference protocol (REF), and two lower dose interleaved protocols (LD1 and LD2).

The linear regression analysis found a high coefficient of determination between US derived velocity measurements and qDSA in all protocols (n = 40, $R^2\hspace{0.17em}\ge \hspace{0.17em}0.93$). Table 3 shows individual values for linear regression and Bland-Altman analyses for each protocol. The lowest $R^2$ of 0.93 was observed for the lowest dose protocol (LD2). All fits include a slope of one and an offset of zero in the 95% confidence interval.

Table 3:

Linear regression and Bland-Altman analyses comparing ultrasound and qDSA derived velocity [cm/s], (n = 40). Results shown for a reference protocol (REF) and two lower dose interleaved protocols (LD1 and LD2). Dose rates cited at the interventional reference point.

Protocol (Dose rate)	Linear Regression			Bland-Altman
	$R^2$	Slope (95% CI)	Offset [cm/s] (95% CI)	95% LOA [cm/s]	Bias [cm/s]
REF ($\mathrm{11}\mathrm{.4 }mGy/s$)	0.97	1.05 (0.98, 1.11)	2.09 (−0.07, 4.25)	[0.65, 6.68]	3.67
LD1 ($3\mathrm{.7 }mGy/s$)	0.96	1.05 (0.98, 1.12)	1.94 (−0.43, 4.30)	[0.33, 6.93]	3.63
LD2 ($1\mathrm{.7 }mGy/s$)	0.93	0.98 (0.90,1.07)	2.02 (−0.96, 5.00)	[−2.52, 5.48]	1.48

The LD1 and REF protocol had very similar Bland-Altman biases and ranges of agreement with ultrasound, whereas the LD2 protocol resulted in a 2.2 cm/s decrease in the bias and a 2.0 cm/s increase in the range of agreement compared to LD1 and REF protocols. Figure 8 shows Bland-Altman and linear regression plots comparing US and qDSA performed on the REF, LD1 and LD2 protocols.

Figure 8:

Top) Linear regression analysis comparing US and qDSA derived velocities using imaging protocols A) REF, B) interleaved lower dose LD1 and C) interleaved lower dose LD2. Solid line shows linear fit. Bottom) Bland-Altman analysis comparing US and qDSA derived velocities. Solid line shows the bias with dashed lines representing limits of agreement (95% confidence interval).

4. Discussion

An interleaved x-ray imaging technique has been developed for simultaneous quantitative measurement of blood velocity (qDSA) and conventional high quality diagnostic DSA imaging. The proposed approach of inserting low-dose, high-rate image frames between full dose frames acquired at a conventional rate improves clinical translatability of qDSA in two ways. First, there is a large dose rate reduction relative to imaging at a high frame rate with all full dose frames. In this study, interleaved protocols with −67.7% and −85.5% dose rate reduction were demonstrated. The protocol providing −85.5% reduction had a dose rate equivalent to conventional DSA without interleaving at (1 – 0.855) x 25.4 fps = 3.7 fps, which is similar to the frame rates commonly employed for DSA (1 – 4 fps).^26,27 Second, the interleaved protocol avoids the need to acquire two scans at different frame rates and different dose levels (e.g. 30 fps at fluoroscopic dose for qDSA, followed by 3 fps DSA for vessel morphology). Avoiding two scans reduces iodine contrast dose and procedure time, which is important for procedures involving multiple DSA acquisitions.

The qDSA velocities derived from reference and interleaved imaging protocols were compared to gold standard US derived velocities via linear regression analysis. Results demonstrated a strong linear relationship to US velocity in all three protocols ($R^2\hspace{0.17em}\ge \hspace{0.17em}0.93$) with near-unity slopes. Bland-Altman analysis found small qDSA velocity biases in the range of 1.48 – 3.67 cm/s (4–11% of mean ambient flow). This is similar to the $\pm 4\%$ accuracy for the ultrasound flow probe (6PXL Transonic), which converts to $\pm 1.8$ and $\pm 1.0$ cm/s for the highest and lowest flow states used in this study. Since the bias was not specific to the use of an interleaved protocol it was not investigated further in the present work. However, additional theoretical and experimental work should be performed to identify all potential sources of bias and variance in the qDSA technique.

The radiation dose rate at the interventional reference point was used to determine dose reduction and evaluate the consistency of dose delivery in the interleaved protocols. The small inter-scan %CV in kV and radiation dose demonstrated consistency across repeated scans with the same protocol. Mild intra-scan variability of kV was observed in the low dose pulses of the LD2 protocol (e.g. a drop from $~$74 to 68 kV over the course of 9 pulses). Although this variation is not ideal, the low inter-scan %CV indicates that the pattern of variation is repeatable from scan to scan, which suggests that a correction may be possible. This is left for future investigation.

There are several limitations to the current study. The feasibility of interleaved x-ray imaging was investigated on a system equipped with research prototype software which is not available commercially. The software was capable of frame-by-frame modulation of x-ray tube output during gantry rotation, whereas the interleaved x-ray imaging technique in its most basic form is envisioned for 2D imaging without rotation. Although the current prototype has operating constraints, no fundamental physical barriers are anticipated for interleaved imaging in a true 2D mode. To allow study of interleaved imaging without the confounding factor of rotation, a relatively simple circularly symmetric phantom was used.

The phantom used in this initial study had a diameter corresponding to the lower end of the range of typical patient thicknesses. Thicker phantoms will demand higher x-ray output and different tube techniques. Generally, an interleaved protocol for a given phantom thickness should have full dose pulses matched to what is selected by the automatic exposure control of the x-ray system for a conventional DSA protocol, and then the chosen ratio of doses for low vs. full dose pulses determines the percent dose reduction. Consequently, a thicker phantom will require higher tube output for both full and low dose pulses, but a percent dose reduction similar to that reported in this work should be obtained if the ratio of doses for low vs. full dose pulses is preserved. Performance of a low dose interleaved protocol under these conditions remains to be investigated.

The qDSA data analysis method used in this work has been investigated for full dose sequences in a porcine model for velocities ranging from 2 to 74 cm/s.³⁰ The range of native fluid velocities (i.e. not including contrast injection flow) was 19 to 41 cm/s in the present study. This range encompasses the typical blood velocities of the human left and right hepatic arteries, which average 28 and 35 cm/s, respectively.³¹ Lower velocities may be encountered depending on clinical scenario. For example, hepatic embolization is usually performed beyond the left and right hepatic arteries,³² and injection of embolic materials also decreases arterial velocity. On the other hand, underlying diseased states (i.e. hepatocellular carcinoma and liver cirrhosis) have both been found to increase hepatic arterial velocity.^33–36 Future work should confirm results in the full range of fluid velocities expected for a clinical scenario.

The simple phantom design did not include vessel motion. Motion is expected clinically, and a motion-compensated qDSA²⁵ algorithm has been developed, but the performance of motion-compensated interleaved qDSA requires further study. Finally, this study investigated only two potential interleaving protocols. This should be followed by a more thorough investigation of possible protocols, including protocols with different full dose pulse rates and different levels of dose rate reduction.

5. Conclusion

The proposed interleaved method uses simultaneous acquisition of low-dose high-rate images for intra-procedural blood velocity quantification (qDSA) and full-dose low-rate images for visualization of vessel morphology. In this study, interleaving low- and full-dose images resulted in up to −85.5% dose rate reduction in qDSA compared to qDSA using all full dose pulses. The protocol providing −85.5% reduction had a dose rate equivalent to conventional DSA without interleaving at 3.7 fps. Furthermore, a comparison of qDSA velocity results to gold standard US measurements demonstrated the LD2 images retained the information needed to quantify velocity in a simple flow phantom. Interleaved x-ray imaging can facilitate clinical translation of qDSA techniques for blood velocity quantification by incorporating the necessary high-rate imaging into a standard low frame rate DSA protocol, while maintaining typical dose levels.