Cardiac three-dimensional rotational angiography can be performed with low radiation dose while preserving image quality

Abstract

Aims

The effective radiation dose (ED) of three-dimensional rotational angiography (3DRA) is 5–8 mSv, leading to reticence on its use. We evaluated the potential of 3DRA with a reduced number of frames (RNF) and a reduced dose per frame.

Methods and results

Three-dimensional rotational angiography was performed in 60 patients (52.5 ± 9.6 years, 16 females) referred for ablation in the right (RA; n = 10) and left atrium (LA; n = 50). In a simulation group (n = 20), the effect of dropping frames from a conventional 248 frames 3DRA LA acquisition was simulated. In a prospective group (n = 40), RNF 3DRA were acquired of LA (n = 30) and RA (n = 10) with 67 frames (0.24 Gy/frame) and 45 frames (0.12 μGy/frame), respectively. Accuracy was evaluated qualitatively and quantitatively. Effective radiation dose was determined by Monte Carlo simulation on every frame. In the simulation group, surface errors increased minimally and non-significantly when reducing frames from 248 to 124, 83, 62, 50, 42, and 31: 0.49 ± 0.51, 0.52 ± 0.46, 0.61 ± 0.49, 0.62 ± 0.47, 0.71 ± 0.48, and 0.81 ± 0.47 mm, respectively (Pearson coefficient 0.20). All 3D LA images were clinically useful, even with only 31 frames. In the prospective group, good or optimal 3D image quality was achieved in 80% of LA and all of RA reconstructions. These accurate models were obtained with ED of 2.6 ± 0.4 mSv for LA and 1.2 ± 0.5 mSv for RA.

Conclusion

Three-dimensional rotational angiography is possible with a significant reduction in ED (to the level of prospectively gated cardiac computed X-ray tomography) without compromising image quality. Low-dose 3DRA could become the preferred online 3D imaging modality for pulmonary vein isolation and other anatomy-dependent ablations.

Introduction

Comprehensive anatomical visualization is essential during ablation of many cardiac arrhythmias. This is especially true for pulmonary vein isolation (PVI), where the anatomically determined region surrounding the pulmonary vein (PV) is targeted for energy delivery, but also applies for other arrhythmia substrates. Anatomical reconstruction by electroanatomical mapping is often inaccurate. Therefore, merging with highly detailed three-dimensional (3D) patient-specific anatomical models is often sought.^1–3 Also overlay of classical fluoroscopic imaging with 3D models has been described as an approach to assist ablation: the fine details about the PV orifices and the ridge between left atrial (LA) appendix and the left PVs, allows for accurate planning and execution of the ablation lines.⁴

Three-dimensional imaging of the cardiac chambers can be performed pre-interventionally by magnetic resonance imaging^5,6 or computed X-ray tomography (CT).^4,7 More recently, also 3D rotational angiography (3DRA) has been introduced, enabling 3D imaging during the intervention which offers advantages in terms of patient comfort, logistics, and accuracy of registration.^1,8,9 Nevertheless, its effective radiation dose (ED) is estimated between 5 and 8 mSv.¹⁰ Although the ED is lower than with retrospectively gated CT (ED of 13.8 mSv)¹¹ and comparable with prospectively gated cardiac CT (ED around 4.5 mSv),¹¹ it is higher than prospectively gated, high-pitch cardiac CT (ED of 0.9 mSv).¹² Since this radiation dose has to be added to the radiation during the ablation itself, some consider this extra radiation dose undefendable.

We wanted to explore ways to reduce the radiation dose of 3DRA while preserving its high anatomical detail during imaging the left or right atrium (LA; RA). We evaluated the impact (i) of a reduced number of frames (RNF) and (ii) of a reduced detector entrance dose per frame on image quality of 3DRA and calculated the respective patient EDs.

Methods

Study population and design

Three-dimensional rotational angiography was performed in 60 patients (52.5 ± 9.6 years, 16 females) referred for ablation of arrhythmias in the RA (n = 10) or LA (n = 50). Mean patient height, weight, and body mass index (BMI) were 177 ± 9.8 cm, 82.6 ± 14.6 kg, and 26.1 ± 3.7 kg/m², respectively. Patients comprised two groups: a simulation group (n = 20) and a prospective group (n = 40).

In the simulation group, the effect of dropping an increasing number of frames from a standard 248 frames 3DRA LA acquisition was simulated: the resulting LA surface models were qualitatively and quantitatively compared with the original 3DRA, and ED was estimated for each simulated sequence. Reconstructed images were segmented using our in-house developed segmentation software (EPSegmenter), and ED was determined by Monte Carlo simulation as reported before by our group.^10,13 After the feasibility of the RNF 3DRA was proven in this cohort, a prospective study was initiated in which 3DRA were performed with a RNF.

In the prospective group, RNF 3DRA protocols were acquired with acquisition parameters closely matching the simulated settings from the simulation study that seemed to preserve imaging quality with a much lower ED. More specifically, RNF 3DRA acquisition protocols with 67 and 45 frames were installed on the system and used for the acquisition of LA (n = 30) and RA (n = 10) 3DRA.

Three-dimensional rotational angiography and surface model generation

All 3DRA in this study were acquired on a floor-mounted Siemens Axiom Artis dBC biplane fluoroscopy system (Siemens) with a large-area 30 × 40 cm flat-panel Si-detector using CsI as the scintillator material. Three-dimensional image reconstruction was performed using Siemens syngo^® DynaCT Cardiac software on a Leonardo workstation.

Fluoroscopic images were sequentially acquired over the course of a 200° rotation of the X-ray tube around the patient, taking 4.13 s to complete. Tube rotation starts at 10° above the horizontal plane and rotates around the backside of the patient to end at 10° above the horizontal plane on the other side of the patient. There is no oblique inclination of the rotation plane with respect to the longitudinal patient axis. The chamber of interest (LA or RA) was positioned in the centre of rotation using two orthogonal single fluoroscopy angles for guidance. Manual collimation of the tube radiation field was performed before starting the imaging run, optimized to eliminate unnecessary radiation of areas cranially and caudally to the LA or RA.

The standard 3DRA imaging protocol with this system is performed at a default dose setting of 0.54 μGy/frame and with acquisition of 248 frames, acquired at 60 frames/s. During earlier work, we had already noted that in patients with lower BMI, lower dose settings per frame seemed feasible.^10,13 Therefore, in the simulation group, a 248 projections (60 frames/s) 3DRA of the LA with target detector entrance doses of 0.24 (n = 15), 0.36 (n = 1), or 0.54 μGy/frame (n = 4) was obtained. For the simulation part of this study, a computer-generated drop of the number of frames was performed offline. Datasets with 124, 82, 62, 50, 42, and 31 frames were reconstructed, segmented, and evaluated in each patient (Figure 1).

In the prospective group, RNF 3DRA was directly acquired at 0.24 μGy/frame with 67 frames (16 frames/s) for the LA in 30 patients, and at 0.12 μGy/frame with 45 frames (10 frames/s) for the RA in another 10 patients.

Contrast was administered directly into the chamber of interest during joint rapid atrial and ventricular pacing at 250 ms. Iomeron 350 contrast (Bracco) was diluted up to 50% using normal saline. Ninety millilitres of diluted contrast was injected at a rate of 20 mL/s, starting 4 s before the actual start of the C-arm rotation.

Image quality evaluation and radiation dose calculation

Qualitative evaluation was performed by grading of the individual reconstructed surfaces following the scheme from Table 1, which has been used before by our group and others to compare LA image quality.^4,9,10,14 The evaluation criteria comprise overall image quality, noise, artefacts, and visualization of specific anatomical structures like the PVs, the PV ostia, anterior ridge, and the LA body (for LA images) and the oval fossa, orifice of both caval veins, Eustachian ridge, and coronary sinus ostium (for RA images). All images were evaluated by two independent readers, blinded to the acquisition protocol/simulation. Their agreement was 81% with a mismatch that was never >1. For non-agreement scores, both agreed on a score while simultaneously looking at the 3D models.

Quantitative evaluation was performed by comparing the surface reconstructions from a RNF 3DRA to the surface reconstruction of the same patient based on the reference acquisition with 248 frames. Eight circumferential points around each PV ostium (8 × 4, simulating ablation of each vein individually) and another eight points around ipsilateral PV ostia (8 × 2, simulating 2 × 2 ipsilateral ablation circles) were annotated on the reference models (Figure 2). A quantitative error was determined by computing the average distances between the segmented surfaces from RNF 3DRA and the points marked on the reference model, either in an 8 × 4 circumferential shape around each PV ostium, or as 8 × 2 points marked around ipsilateral PV ostia. For each point, the distance to the RNF 3DRA surface model was computed.

Radiation dose was measured using calibrated air-ionization chamber dose–area product (DAP) metres, incorporated in the tube housing. Both total DAP for every imaging run and DAP values for individual frames were recorded. The ED was simulated based on the frame-by-frame DAP measurements by Monte Carlo simulation with the PCXMC software (STUK), as previously described.^10,13 Total ED of a 3DRA run was computed by summation of the ED of all frames. For the simulation group, only the dose of the projection images that were used to obtain a simulated image was taken into account. Hence, this calculation assumes that the imaging system X-ray auto-exposure routine predicts nearly identical X-ray tube levels compared with the reference acquisition when the projection images are spaced with an increasing angular increment. In the prospective group and the reference acquisition, the ED of all frames was summated into a total ED.

Results

Simulation group

The quantitative assessment of the distance error for points around the PV ostia for each simulated RNF reconstruction vs. the reference LA reconstruction is shown in Figure 2. A histogram of the averaged result for all patients and all RNF reconstructions is presented in Figure 3. Analysis of variance showed no difference between the reconstructions (P = 0.28) although a small increase in reconstruction error with a smaller number of projection frames is perceived visually (Figure 3): the difference with the points on the reference model was 0.49 ± 0.51, 0.52 ± 0.46, 0.61 ± 0.49, 0.62 ± 0.47, 0.71 ± 0.48, and 0.81 ± 0.47 mm in the 124, 83, 62, 50, 42, and 31 frames models, respectively (Pearson coefficient = 0.20). But even with the lowest numbers of frames (i.e. 31), the average distance error remained <1 mm, with 95% of differences ≤2.5 mm and 99% ≤4.7 mm, which is still within acceptable margins for ablation purposes.

Also based on a qualitative evaluation of the reconstructed surfaces in the simulation group, none was considered non-diagnostic as shown in Figure 4, left panel. The majority of the RNF LA surfaces were scored as good or optimal, even when reconstruction was performed with only 31 frames vs. the standard of 248.

The ED for the reference 3DRA and for the RNF 3DRA are shown in the left panel of Figure 5. As we assumed an identical behaviour in terms of auto-exposure control and per frame DAP values vary continuously, the RNSs translates in a linear ED reduction.

Prospective group

A RNF 3DRA with 67 frames at 0.24 μGy/frame was used for imaging the LA. This mode was chosen because estimated average distance error from the simulation experiment was below 0.65 mm with a predicted dose reduction of four-fold (cf. Figure 5), while still preserving a margin of safety for quality (since in case of a non-useful 3DRA, a second acquisition would be necessary resulting in a extra radiation dose). Imaging of the RA was performed with 45 frames at 0.12 μGy/frame because less detailed anatomy is generally needed than for the LA.

Good or optimal 3D image quality could be achieved in 80% of the LA reconstructions (24/30) and all of the RA reconstructions (Figure 4, right panel), i.e. all the 3D reconstructions were considered perfectly useful for accurate anatomical ablation target guidance. Examples of good-quality images of LA and RA are shown in Figure 6. They compare well with 248 frame models with higher entrance dose rates that we published previously^9,13 and to the reference models from the simulation group (Figure 4, left panel, 248 frames).

The 3DRA acquisitions for the LA and RA surface models were performed with an ED of 2.6 ± 0.4 mSv for the LA and 1.2 ± 0.5 mSv for the RA (Figure 5, right panel).

Discussion

To our knowledge, this is the first study reporting on low-dose 3DRA imaging, based on a reduction of the projection frames and a reduced dose per frame. Our data show that a major reduction of the ED can be obtained by this technique while maintaining an equivalent image quality that is clinically perfectly useful. The average surface errors from the majority of the simulated RNF 3DRA compared with the reference model, based on 248 frames, were ≤1 mm. Implementation of the RNF imaging in the prospective group confirmed that image quality could be preserved compared with conventional 248 frames 3DRA. Although not compared in a randomized fashion, ‘optimal’ quality was achieved in up to 36% using conventional 3DRA in prior series^9,13 and 40% in the simulation group of this study, vs. 33% using low-dose 3DRA. ‘Good’ image quality was achieved in 29–67% with conventional vs. 47% with low-dose 3DRA. Whether 20% with only ‘useful’ imaging during RNF 3DRA could be improved by a higher number of frames can only be evaluated by performing both types of 3DRA in the same patient, which has ethical concerns due to the cumulative radiation dose for such study. Anyway, the image quality of conventional cardiac 3DRA has been shown to be comparable with that of cardiac CT⁹ and implementation of RNF 3DRA seems to adequately preserve that image accuracy.

The preserved anatomic accuracy could be achieved with a reduction of ED by a factor 4–5. The ED dose of a pre-procedural conventional prospectively gated cardiac CT is ±4.5 mSv and is comparable with those of per-procedural conventional cardiac 3DRA, i.e. ±6.9 mSv.^{11,13,15–17} These doses are substantially higher than the currently achieved doses with the RNF 3DRA, i.e. 2.6 ± 0.4 mSv for the LA and 1.2 ± 0.5 mSv for the RA. RNF 3DRA can also compete in terms of ED with prospectively gated high-pitch cardiac CT, which is only available on the most recent multislice CT scanners.^12,18,19

The fact that 3DRA is performed during the ablation procedure has additional advantages: no prior imaging means easier planning and logistics, and per-procedural 3D imaging can improve registration due to differences in filling status, heart rhythm, intra-abdominal pressure, etc. Anatomic accuracy is of utmost importance for effective and safe ablation. Therefore, we believe rotational angiography has an edge on pre-procedural CT imaging.

When comparing the ED in the prospective group with the post hoc simulation group, it should be noted that the predicted linear dose reduction is not completely reached. This can be explained by the non-linear behaviour of the auto-exposure algorithm that determines optimal image radiation parameters for each projection image of the 3DRA scan based on the previously acquired image. With a lower number of frames, the algorithm detects a decrease in the overall X-ray tube load and decides to increase the frame dose to realize projection images of higher quality. In other terms, the loss in number of frames comes with an increase in the quality of the individual images, at the cost of increasing the ED slightly. This suggests that even further dose reduction may be possible by further lowering dose-per-frame settings or a further reduction in number of frames. The fact that we already achieved highly qualitative 3D images of the RA in this study with only 45 frames at 0.12 μGy/frame underscores this potential.

The low ED of RNF 3DRA imaging (much smaller than a coronary angiography),²⁰ but with the benefit of having a patient-specific 3D image may also lead to its use in ‘simpler’ ablation procedures (like for right atrial flutter, atrioventricular nodal reentrant tachycardia, ectopic atrial tachycardia, right ventricular outflow tract tachycardia, or accessory pathways), where 3D imaging is now deemed undefendable because of the associated radiation or of the cost of non-fluoroscopic imaging. Catheter ablation procedures, in which previously a 2D angiography, may have been performed (e.g. flutter RVOT or some accessory pathways) may also benefit from RNF 3DRA acquisition. Surface models can assist target planning and 3D mapping of potentials or activation times. One could hypothesize that in some of these cases, the added radiation dose of RNF 3DRA might be compensated by a shorter procedure with less fluoroscopy. This area can now be conceived as realistic for prospective clinical evaluation.

Conclusions and clinical implications

Through a reduction in the number of acquired projection frames and a reduction in detector entrance dose, 3DRA can be realized with a significant reduction in ED without compromising image quality. The ED is reduced to the level of state-of-the-art prospectively gated cardiac CT imaging protocols. Using such low-dose protocols is of benefit to all patients in whom a 3DRA is done, and should be recommended to improve long-term safety. Our analysis suggests that even further dose reductions may be possible.

Moreover, 3DRA has other advantages; (i) the workflow is simpler (no need to send the patient to the radiology ward before, and transferring images to the electrophysiology (EP) system); (ii) registration of the 3D image is more accurate (since acquired under the same conditions as the EP study) and simpler (especially if the 3D image is integrated with the real-time fluoroscopy or with a non-fluoroscopic imaging system that is tied to the X-ray equipment like MediGuide); and (iii) the added cost is likely lower (since no other hardware than the already available X-ray system is needed and many modern interventional X-ray systems contain rotational capabilities standard or as an option). Therefore, RNF 3DRA could become the preferred online 3D imaging modality for PVI and other anatomy-dependent ablation procedures, even for procedures where 3D imaging is not performed today.

Acknowledgements

The authors would like to thank Christoph Koehler for his aid in reconstructing the simulated datasets and providing feedback about the auto-exposure parameters used in Siemens SyngoTM DynaCT acquisition.

Conflict of interest: H.H. is the holder of the AstraZeneca Chair in Cardiac Electrophysiology, University of Leuven. H.H. is a member of the scientific advisory board of Siemens Medical Solutions, and receives unconditional research grants through the University of Leuven from St Jude Medical, Medtronic, Biotronik, and Boston Scientific Inc. J.E., S.D.B., and H.H. received funding for related research through the University of Leuven from Siemens Medical Solutions.

References

Author notes