Optimising image quality in intravenous cerebral cone beam computed tomography

Abstract

Introduction

The efficacy of intravenous cerebral Cone Beam Computed Tomography (IV CBCT) is well established; however, image quality has only ever been authenticated by subjective evaluation. The aim of this study was to quantify the factors pertinent to achieving consistent and optimal image quality when performing IV CBCT.

Methods

Between 1 March 2021 and 30 October 2022, 79 patients received IV CBCT. These candidates were divided into three main acquisition field size categories (22/32, 42 and 48 cm) according to the clinical indication. The images were analysed using both a quantitative assessment and a subjective evaluation. Here, a comparison of Hounsfield units (HUs), signal‐to‐noise ratio (SNR), contrast‐to‐noise ratio (CNR) and noise index was conducted for each study and compared relative to the acquisition field size. The subjective analysis was performed in a non‐blinded fashion where the diagnostic value (DV) of the exam was determined according to a graded scale. A phantom analysis for each of the acquisition field sizes was conducted and modulation transfer function (MTF) graphed.

Results

Significantly higher HU, SNR, CNR and lower noise indices were achieved with the 42‐cm protocol than the 22/32 and 48‐cm protocols. Here a greater DV was also reported. The MTF demonstrates marginally improved spatial resolution for the 22‐cm protocol, but this is near equivocal for the 32‐, 42 and 48‐cm protocols.

Conclusion

The use of larger acquisition field sizes provides improved image quality when performing IV CBCT as an alternative to intra‐arterial (IA) CBCT.

The efficacy of intravenous (IV) Cone Beam Computed Tomography (CBCT) is well established; however the image quality has only ever been authenticated by subjective evaluation. The aim of this study was to quantify the factors pertinent to achieving consistent and optimal image quality when performing IV CBCT.

Introduction

Intracranial endovascular treatments warrant rigorous follow‐up imaging. This is especially true for treatments that include the deployment of an intracranial endovascular stent as reported in‐stent restenosis rates of 8–30% exist. ¹ Use of flow diverter stents for unruptured aneurysms has increased in Australia ² and hence a meticulous follow‐up protocol is essential to assess stent position, including wall apposition, as well as early detection of in‐stent stenosis, all while surveilling the aneurysm regression. The primary imaging modalities of magnetic resonance imaging (MRI) and computed tomography (CT) are reported as being subject to significant artefacts that inhibit the ability to assess true luminal capacity of cerebral stents including, but not limited to, beam hardening artefacts and insufficient spatial resolution. ¹ , ³ , ⁴ , ⁵ , ⁶ Cone beam CTs performed on a digital angiography unit have long been regarded as the gold standard for cerebral vascular imaging combining superior spatial resolution to other imaging modalities ² as well as an assessment of flow dynamics, where necessitated. Therefore, intra‐arterial cone beam CT (IA CBCT) in conjunction with cerebral digital subtraction angiography (DSA), an invasive technique with a permanent neurological complication rate of 0.5–0.8%, ¹ is often performed as a first‐line investigation.

While cone beam CTs performed with intra‐arterially injected contrast media result in greater enhancement of iodine due to increased iodine concentration within the relevant vasculature, ⁴ , ⁵ there remains significant cost and risk to the patient. Intravenous CBCT (IV CBCT) is a viable non‐invasive alternative that has largely been accepted by interventional neuro‐radiologists as it can provide equivalent spatial resolution to IA CBCT, has inherent low risk as well as concurrent opacification of all intracranial vascular territories. ⁶

While the efficacy of IV CBCT is well established, ¹ , ³ , ⁶ , ⁷ , ⁸ , ⁹ our centre felt that further refinement of the established protocol was warranted. Despite being of good diagnostic value (DV), our initial studies were somewhat subject to substantial image noise and poor iodine enhancement. Our aim was to achieve an image quality that was not only superior to these preliminary studies, but also consistent and reproducible. Additionally, to the best of our knowledge, literature evaluation of image quality in IV CBCT has only been authenticated by subjective assessment. ¹ , ³ , ⁴ , ⁶ , ⁷ , ⁸

Methods

Ethics approval

This retrospective quality control study was approved by The Alfred Hospital's research organisation Project number 220/23. Our hospital is a Level 1 trauma centre and a tertiary referral centre in Australia.

Inclusion criteria

Included in the review were all patients aged 18–99 who underwent an IV CBCT performed at our institution from 1 March 2021 to 30 October 2022. One patient was excluded from the statistical analysis due to non‐compliance with instructions and gross movement during the acquisition.

Patient demographics

There were 79 patients who received an IV CBCT. This included 33 males (42%) and 46 females (58%) with an age range of 22–84 years. The primary indications for imaging were follow‐up of flow diverter implantation (n = 32, 40.5%), anatomical mapping prior to stereoelectroencephalography (SEEG) (n = 23, 29.1%), follow‐up of aneurysm coiling (n = 11, 13.9%) and characterisation of aneurysms or arteriovenous fistulas (n = 13, 16.5%).

Imaging technique

In all cases, the candidates were consented for the technique and use of iodinated contrast media before being positioned supine with their head in a carbon fibre holder. An immobilisation strap was placed over the candidate's head. An 18 gauge cannula was inserted into the median cubital vein and 95 mL of iohexol (Omnipaque, 350 mg/mL, GE Healthcare, Piscataway, New Jersey) followed by a 50‐mL saline chase was injected at 8 mL/sec using a powered injector. Scans were performed using a 10‐sec dual acquisition (Artis Q Biplane, Siemens AG, Forchheim, Germany) with the parameters of 200° angular arc, 1.2 μGy/frame, 0.4 mm focal spot, 70 kV and automatic dose rate control (ADRC). Visual bolus tracking was used to determine the appropriate timing of the contrast‐enhanced sequence.

The acquisition field of view (FOV) was modified according to the primary indication for follow‐up where patients were divided into three main categories. Of the 79 candidates, 15 had scans acquired with a 22/32 cm FOV (18.99%), 30 with a 42 cm (37.97%) and 34 with a 48 cm (43.04%). For the purpose of statistical analysis, the patients who received scans using either a 22 or a 32 cm FOV were combined into a single broad category.

The rotational volumes automatically transfer to a dedicated workstation (Syngo XWP VD20C; Siemens, Erlangen, Germany) for post processing. As per standard practice at our institution, these were reconstructed as multiplanar reconstructions (MPRs) using maximum intensity projection (MIP) processing in both full field and refined volumes using an image matrix of 512 × 512. Additionally, the volume was displayed as a volume rendered 3D model. Motion correction was applied where required and processing algorithms of smooth or sharp applied, where indicated, to enhance the anatomy relevant to the clinical question.

Quantitative analysis

A comparison of Hounsfield units (HUs), signal‐to‐noise ratio (SNR), contrast‐to‐noise ratio (CNR) and noise index was conducted for each study and compared relative to the acquisition field size.

Measurements were obtained by one observer (L.B.) at the workstation. All series were refined to a 9‐cm FOV. HU were obtained by placing an elliptical region of interest (ROI) within the terminal internal carotid artery (ICA), maximally sized to the lumen of the artery. Image noise was defined as the standard deviation of the attenuation of the background (SD^b). The SNR was calculated using the formula SNR = HU^ICA/SD^b. The CNR was calculated by CNR = (ROI_a – ROI_b)/SD_b, where ROI_a and ROI_b were the HU of the terminal ICA and background, respectively. A 5‐mm ROI in the temporal lobe immediately adjacent to the terminal ICA was used to define the background.

Subjective image analysis

A subjective analysis of the reconstructed images was performed in a non‐blinded fashion allowing for assessment of the clinical DV of the exam. This was graded on a scale of 0–2 with 0 being overall non‐diagnostic, 1 being diagnostic for the target but with deficits in the evaluation of non‐target intracranial vascular territory and 2 being diagnostic for the whole intracranial territory. The patients receiving IV CBCT for anatomical mapping prior to SEEG were uniformly assessed as DV2.

Phantom analysis

CBCT images were acquired using the resolution and geometry test module of a Catphan 604 image quality phantom (The Phantom Laboratory, Salem, NY). Scans were acquired using an identical 10‐sec acquisition protocol as utilised throughout the study (200°, 1.2 μGy/frame, 0.4 mm focal spot and ADRC). A set kV of 125 was used in conjunction with small focal spot size forcing the ADRC to maintain the dose to the detector via mA rise and pulse width as opposed to varying the kV. Field sizes of 22, 32, 42 and 48 cm were obtained. The images were then transferred to the dedicated workstation and reconstructed to ‘refined volume’ 9‐cm FOVs with an image matrix of 512 × 512.

A 50‐μm embedded steel wire within the Catphan phantom was used to determine the modulation transfer function (MTF) of each reconstructed image set. As the steel wire is embedded in the phantom at an angle of 5°, slice profiles were taken across the wire on consecutive axial slices and a translation correction applied to create a supersampled point spread profile. A Gaussian function was fitted to the supersampled point spread profile and MTF calculated from this function via fast Fourier transform (FFT) using the Data Analysis ToolPak in Microsoft Excel (Microsoft Corp., Redmond, WA).

Statistical analysis

Statistical analysis was performed using IBM SPSS Statistics for Windows, version 29.0.0.0 (IBM Corp., Armonk, NY). A one‐way ANOVA with Scheffè's correction was used to identify the statistical significance between the image acquisition parameters and quality measurements with respect to field size. Comparisons between patient demographics were analysed using a Pearson's chi‐square analysis. P < 0.05 was considered to indicate statistical significance. All data are expressed as mean ± SD unless otherwise specified.

Results

An analysis of the studies obtained early in the trial indicated that despite a target beam energy setting of 70 kV, ADRC resulted in an average beam energy substantially higher when the CBCTs were obtained utilising small field sizes (Table 1). This was attributed to the positioning of the ADRC with greater predominance over dense osseous regions and kV ramping; a consequence of anode heat loading and mA output, related to quantum mottle repression. The subsequent series were obtained using larger field sizes (42–48 cm) with the ADRC positioned mid‐brain as this resulted in subjectively improved image quality. This formulated the basis for our quantitative analysis.

Table 1.

Mean demographics and imaging parameters for intravenous cerebral Cone Beam Computed Tomography acquired using different field sizes.

Field size (cm)	Age (years)	BMI	Average kV ± SD	mA	Air kerma (mGy)	DAP (μGy × cm2)	Fluoroscopy time (s)	Acquisition image matrix	Pixel binning
22/32	49.5 ± 8.8	25.8 ± 2.4	90.71 ± 8.8	337.7 ± 26.7	332.28 ± 22.94	2579.0 ± 1042.3	8 ± 5	1024 × 1024 (22) 720 × 720 (32)	Non‐binned 2 × 2
42	58.53 ± 12.29	29.84 ± 6.1	75.14 ± 1.9	387.3 ± 20.7	292.0 ± 19.20	5846.69 ± 414.94	10.3 ± 4.7	960 × 960	2 × 2
48	44.1 ± 17.17	26.58 ± 5.3	76.09 ± 2.2	389.55 ± 10.19	289.87 ± 28.2	7550.87 ± 763.96	9 ± 3	1240 × 960	2 × 2

There was no statistical variance in patient demographics between the three field size cohorts (P > 0.05).

The analysis of the image acquisition parameters with respect to field size is characterised in Table 1. Quantitative assessment shows a substantially lower average kV (P < 0.001) for sequences acquired using field sizes greater than 32 cm but no statistical variation between the 42 and 48‐cm protocols (P = 0.863). While there is no statistical variation in mA between the three protocols, significance was identified when comparing air kerma between the larger and smaller field protocols (P < 0.001).

Analysis also demonstrates higher dose area product (DAP) (P < 0.001) for sequences obtained using the 42 and 48 cm field sizes, representing an overall increase of 127% and 193%, respectively.

Table 2 summarises the quantitative analysis for image quality relative to field size. The mean HU of the terminal ICA was measured at 1047.03 ± 229 for the 42‐cm protocol. This is substantially higher than the 765.7 ± 108 for the 22/32‐cm protocols (P = 0.01). The 42‐cm protocol also demonstrates a 128% increase in SNR and a 142% increase in CNR when compared to those of the smaller field of views (P < 0.001 and P < 0.001 respectively).

Table 2.

Image quality characteristics according to field size.

Field size (cm)	ICA (HU)	SNR	CNR	Noise index	Mean DV
22/32	765.71 ± 108	8.05 ± 2.63	7.20 ± 2.82	99.92 ± 28.59	0.93
42	1047.03 ± 229	18.36 ± 5.61	17.49 ± 5.59	59.23 ± 11.75	1.57
48	871.4 ± 231.62	13.72 ± 5.20	13.37 ± 5.15	69.03 ± 24.44	1.88

A comparison of the CNR between the 22/32‐ and 48‐cm protocols also demonstrates statistically significant improvement (P = 0.003, 85% increase) without a corresponding increase in SNR (P = 0.08). Statistical significance in quality measurements between the 42 and 48‐cm protocols was also identified (SNR P = 0.007, CNR P = 0.02). While the quantitative analysis shows no significant change in background image noise between the 42 and 48‐cm protocols (P = 0.20), there is a statistically significant increase when these are compared to the 22/32‐cm protocols (P < 0.001).

Subjective analysis indicates greater diagnostic value associated with sequences obtained using larger field sizes. Overall, 13% of the 22/32‐cm protocol studies was deemed undiagnostic (n = 2) and 3% of the 48‐cm protocol (n = 1).

In the interests of identifying other parameters that may influence image quality outside the scope of our research, a literature search through PubMed was conducted. Variations of the following terms were used to identify relevant manuscripts that detailed different IV CBCT techniques. Combinations of the following search terms were used: ‘cone beam CT’, ‘intravenous’, ‘flow diverter’, ‘c‐arm’ and ‘flat detector’. This search revealed a wide variance in image acquisition technique for IV CBCT as summarised in Table 3.

Table 3.

A summary of the different injection and acquisition techniques employed by different authors. ¹ , ³ , ⁴ , ⁶ , ⁷ , ⁸

References	Acquisition type	Acquisition duration per spin phase (sec)	Matrix size	Contrast media concentration (mg/mL)	Contrast injection volume (mL/sec)	Injection rate (mL/sec)
Yu et al. 7	Single	20	512 × 512	350	80	4
Saake et al. 4	Single	20	512 × 512	350	100	5
Toyota et al. 6	Dual	5	256 × 256	350	65	9
Golitz et al. 5	Dual	10	512 × 512	300	60	5
Rosch et al. 3	Dual	5	512 × 512	400	60	5
Buhk et al. 1	Single	20	512 × 512	350	100	5
This study	Dual	10	512 × 512	350	95	8

Discussion

Tube potential and iodine attenuation.

Previous studies definitively state that the HU of iodinated material is greatly enhanced by utilising a low kV acquisition technique to bring the mean photon energy (approximately half the peak tube voltage) closer to the 33.2 keV K‐edge of iodine. ¹⁰ , ¹¹ , ¹² Here the photoelectric effect occurs in greater predominance, resulting in increased attenuation of iodinated materials. ¹¹ , ¹³ , ¹⁴ Similar studies have shown that this can result in an increase in iodine enhancement of up to 70% when comparing CT scans acquired at 80–120 kV. ³ Conversely, this also results in fewer photons striking the detector and thus a greater incidence of image noise and beam hardening artefacts. ¹⁰ , ¹² , ¹³ , ¹⁵

Our study demonstrates a statistically significant increase in the SNR and CNR with an increase in the acquisition field size. This was thought to result from the subsequent reduction in average beam energy directly related to the density of the anatomical structures positioned with prevalence in the centre of the FOV. The highest image quality was achieved with the 42‐cm protocol, with an average beam energy of 75.14 kV, where the SNR and CNR improved by 128% and 142%, respectively when compared to the 22/32‐cm protocol. Contrary to the literature, our study showed a reduction in image noise for the lower kV protocols. Arguably this represents a contributing factor to the inherent improvement in image quality where there is less degradation of signal from quantum mottle as utilising larger field acquisitions results in an, albeit small, reduction of the spatial resolution (Fig. 1).

Figure 1.

Measured and fitted modulation transfer function for 22, 32, 42, 48‐cm fields of view, using the small focus.

As the 22‐cm protocol combines a large acquisition matrix with a non‐binned technique, it yields images with the greatest inherent spatial resolution. Where an assessment of stent fracture or deployment diameter is warranted independent of a concurrent assessment of the relevant vasculature, the use of this field size may offer a proposed benefit. However, our study demonstrates that the use of the 22‐cm FOV is also associated with a 48–69% increase in image noise in conjunction with a considerable reduction in the SNR and CNR.

It is our experience that while phantom analysis shows improved spatial resolution for the smallest acquisition field size (Fig. 1), the resultant degradation of image quality from the substantial increase in image noise combined with systemic need for a higher kV output offsets this benefit (Fig. 2).

Figure 2.

Identical patient – a comparison of 22‐cm versus 42‐cm FOV with regard to acquisition KV, spatial resolution and image noise. (A) A significant increase in in image noise when compared to (B) obscuring visualisation of the orbitofrontal and anterior temporal arteries.

Furthermore, we contend that the ability to reconstruct IV CBCT images to a ‘refined volume’ from larger acquisition field sizes further offsets this handicap as invariably the achieved voxel size exceeds the spatial resolution capacity of the imaging system. ¹⁶ ‘Refine volume’ allows the operator to retrospectively reconstruct the 512 × 512 data set to a smaller field, resulting in a smaller voxel size and higher output spatial resolution (Fig. 3). Moreover, the achieved spatial resolution for 42/48‐cm protocols still exceeds the maximal capacity of modern CT systems. ¹⁷

Figure 3.

‘Refined volume’ reconstruction combined with standard and sharp algorithms. (A) Full skull, 10mm MIP. (B) Full skull, digital zoom, 10mm MIP. (C) 'Refined volume', 10mm MIP. (D) 'Refined volume', sharp algorithm, 10mm MIP.

Throughout the duration of the study period, three subjects received two IV CBCTs as part of routine follow‐up. In each of these cases, the initial IV CBCTs were performed using a 22‐cm FOV while the subsequent studies were performed with a 42‐cm FOV. In all three scenarios, there was a substantial increase in CNR (5.1825 vs. 15.2485) and SNR (6.2479 vs. 16.0491) in the latter protocols.

Imaging technique and contrast injection rate.

The difference in acquisition techniques between our own study and the relevant literature is multifactorial (Table 3). Where other studies have maximised their CNR by using a 20‐sec single spin acquisition in combination with a 5‐mL/sec injection, ¹ , ⁴ , ⁷ , ¹⁸ in our experience the propensity for patient movement, despite immobilisation, is too great for such a long acquisition. Our study also utilises a dual acquisition allowing for the creation of detailed volumed rendered reconstructions, the ability to reformat the data sequence using ‘motion correction’ as well as the production of dedicated subtraction maps necessary for stereotactic surgery planning.

The degree of vascular enhancement is directly proportional to the injection rate of contrast media. ¹⁹ , ²⁰ An early article by Toyota et al. describes an injection rate of 9 mL/sec in combination with a 5‐sec acquisition and a reconstruction matrix of 256 × 256. In this article, the authors necessitate reconstructing the data to a much smaller image matrix to maximise the inherently poor contrast resolution at a substantial cost to spatial resolution.

More recent literature has identified that the proposed benefit of injection rates greater than 8 mL/sec are inconsequential, owing to reflux of iodinated materials in the IVC and hepatic veins, even in patients absent of heart failure. ¹⁹ , ²⁰

As our initial studies were already conducted with an injection rate of 8 mL/sec but were still perceived as being subject to poor iodine enhancement, with reference to this we looked to quantify other parameters within our control for the improvement of image quality.

Where imaging technique is concerned, the article by Golitz et al. ⁵ is the only true comparison to our own study. Here the authors describe an acquisition technique identical to our own in combination with a lower iodine delivery rate, where iodine delivery rate is a product of the injection rate (mL/sec) and contrast concentration (mg/mL). Decisively a numerical assessment of image quality in IV CBCT was not performed for this article, suffice to say it has a specificity for remnant aneurysm detection similar to MRA and could be comparable to IA CBCT. With reference to this, we demonstrate that contrast can be safely administered at 8 mL/sec with the benefit of maximising relevant vascular enhancement. ¹ , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ¹⁹ , ²⁰

DAP

DAP and, when converted, the effective dose (mSv) are relative to the amount of tissue irradiated and reflective of stochastic risk. ²¹ , ²² CBCTs obtained using larger field sizes undeniably result in a higher DAP owing to greater volumes of irradiated tissue (22/33 cm vs. 42 cm; 2579.0 ± 1042.3 vs. 5846.69 ± 414.94).

Tangibly the 42 and 48‐cm protocols in our study correlate through Australian diagnostic reference levels (DRLs) to a cerebral DSA of two and four vessels, respectively. ²³ While our study did not quantify the effective dose for different field sizes, relevant literature has cited an effective dose of 2.9 mSv for full head CBCTs. ⁴ , ⁸ According to the Radiation Protection Series Publication No. 8 ²⁴ this relates to a low risk of radiation‐induced cancer (~10⁻⁴ to 10⁻³) compared to the general population. However, other publications conclude that a clear cancer risk related to effective doses (<100 mSv) remains unquantifiable. ²² , ²⁵ Furthermore, there is increasing traction to indicate that CBCTs obtained as an independent study can be performed at a substantial reduction in effective dose than the same study performed using a modern multislice CT scanner. ²⁶ , ²⁷

Reflectively, in our study there are many ways the DAP could be reduced further without compromising image quality. Although not appropriated in many clinical scenarios, the use of a single spin instead of a dual would halve the DAP and thus the effective dose while retaining the DV. Albeit not proportional to the reduction in irradiated field size, collimation, where appropriate, has also been shown to reduce DAP. ⁵ , ²¹ This however, carries the disadvantage of losing clinically relevant information regarding intracranial vasculature remote to the implant, which may necessitate further examinations.

The effect of these protocol modifications on image quality, particularly collimation which may also alter the beam characteristics, would need to be studied more closely at a later date. Additionally, the benefit of an assessment of many vessels in a singular examination is not to be understated. ⁸

Conclusion

Increasingly the use of IV CBCT is being indicated as a non‐invasive alternative to traditional intra‐arterial cerebral DSA for different clinical scenarios. Our study has found that the key to achieving optimal and consistently reproducible image quality and vascular enhancement is the use of a large acquisition field size to achieve a low kV acquisition in combination with a high iodine delivery rate. Spatial resolution, while inherently reduced using larger FOVs, is greatly enhanced with the application of retrospective, small FOV reconstruction to achieve the smallest possible voxel sizes while simultaneously retaining a high DV for the whole intracranial vasculature.

Conflict of interest statement

We have no conflict of interest to disclose.

Ethics statement

Ethics for this retrospective quality control study was approved by The Alfred Hospital's research organisation Project number 220/23.

Consent statement

Consent for the use of all relevant images was obtained verbally from each patient at the time of imaging. This was documented in the patient's medical history at our institution in accordance with our hospital policy.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.