Image quality comparison of single-energy and dual-energy computed tomography for head and neck patients: a prospective randomized study

A total of 80 patients scheduled for a head and neck CT were included in this prospective, single-center study. The patients’ diseases covered a mixed spectrum including malignant and non-malignant conditions (Table 1); most patients with suspected malignancy underwent head and neck CT due to staging completion. Four patients had a primary malignancy in the head and neck region (Fig. 1). Patients were randomly assigned to either the DECT or the SECT study group prior to examination (40 patients per study group). Subjects with contraindications for CT imaging with intravenous administration of iodine contrast agents, such as history of allergies or intolerance to iodine contrast agents, nephropathy grade 4 or higher (estimated glomerular filtration rate < 30 mL/min/1.73 m²), hyperthyroidism, age < 18 years, and pregnancy, were excluded. All patients signed written informed consent for CT examination and study participation. The study was approved by the local Institutional Review Board and adheres to the Health Insurance Portability and Accountability Act (HIPAA) criteria and the Declaration of Helsinki.

All examinations were performed on a third-generation dual-source CT scanner (Siemens SOMATOM Force, Siemens Healthcare GmbH) equipped with two 192-row energy integrating detectors (Siemens Stellar-Infinity, Siemens Healthcare GmbH) arranged in an orthogonal direction. The scanner provides a maximum of 240 kW generator power (2 × 120 kW).

In both study groups, CT acquisition of the neck was performed with both arms lowered and placed beside the trunk. After performing the localizer and determining the field of view, 80 mL of iodine contrast agent (Imeron 350 mg/mL, Bracco GmbH) was intravenously injected at a flow rate of 3 mL/s, followed by a saline bolus (30 mL, 3 mL/s). Both the DECT and the SECT examination protocols were performed after a fixed delay of 80 s for all study participants. A real-time automatic tube current modulation algorithm (CARE Dose4D, Siemens Healthcare GmbH) was used for all examinations in both study groups.

An automatic TVA algorithm (CARE kV, Siemens Healthcare GmbH) was used for the SECT study group. TVA is capable of tube voltage settings from 70 to 150 kV in discrete steps of 10 kV. A preselection of TVA presets for different applications is available and can be directly adjusted by the user on a 12-point slider. In this study, a mid-level TVA setting (slider position 7 of 12) was chosen to keep the CNR balanced. To avoid excess radiation dose, TVA was limited from 70 to 120 kV.

DECT acquisition was performed with tube voltages of 80 kV (tube A) and 150 kV (tube B). Both x-ray beams were pre-filtered with an aluminum bowtie filter, with tube B equipped with an additional 0.5-mm tin prefiltration in order to refine the spectral separation. Table 2 provides a detailed overview of the acquisition and reconstruction parameters for the SECT and DECT groups.

Based on DECT raw data, WAI series were reconstructed in-line on the scanner console with a slice thickness of 3 mm in axial orientation and soft (Bf40) and sharp (Br64) convolution kernels. In principle, WAIs are intended to provide image quality equivalent to that of conventional SECT reconstructions. A multiband filtered algorithm was applied to DECT raw data as recommended by the vendor (F-type reconstruction 0.7, nonlinearly merging 70% of the 80 kV and 30% of the 150 kV data spectra).

Like the WAIs, SECT data were reconstructed with 3-mm slice thickness in axial orientation using equivalent soft (Bf40) and sharp (Br64) convolution kernels. Convolution kernels for WAI and SECT images were vendor matched to ensure comparability. Concordant to WAI, no iterative reconstruction algorithm was used in order to limit the bias between SECT and DECT images.

After anonymization, evaluation of image quality was performed on a dedicated workstation with a server/client-based software (Syngo.via VB 20, Siemens Healthcare GmbH) by two board-certified radiologists with 6 and 11 years of experience in head and neck imaging. All imaging series were presented in a randomized order, and both readers were blinded to other imaging data and the patients’ medical history. Regions of interest (ROIs) were placed on axial slices in both jugular veins, in both sternocleidomastoid muscles, and in both fat-containing parapharyngeal spaces. All ROIs were manually drawn as large as possible while carefully avoiding the inclusion of neighboring structures and artifacts. Reproducibility was ensured by defining specific anatomic positions on the axial slices: ROIs in jugular vein and sternocleidomastoid muscle were drawn between the lower boundary of the mandible and the hyoid bone, ROIs in the parapharyngeal space were drawn between the angle and the head of the mandible. In both positions, the readers had to check for artifacts before ROI measurements were performed. See Fig. 2 for an illustration of ROI placement. In order to obtain objective parameters of image quality, mean attenuation (A) of every ROI was measured in Hounsfield units (HU). Fatty tissue within the parapharyngeal space was used as the reference for image noise (N) by measuring the standard deviation of the mean attenuation. Image review began with a default soft tissue (center 50 HU/width 400 HU) and bone window (450/1500 HU) that could be adjusted at the reader’s discretion. CNR was calculated using Eq. 1 and dose-normalized CNR (CNRD) was calculated using Eq. 2.

Sixteen anatomical substructures were evaluated using a default soft tissue window setting (center 50 HU/width 400 HU) according to the European Guidelines on Quality Criteria for CT [10]. The substructures analyzed were as follows: pharynx (1: wall of the pharynx, 2: mucosal margins, 3: parapharyngeal fat spaces, 4: parapharyngeal muscles), larynx (5: wall of the larynx, 6: mucosal folds, 7: perimucosal fat spaces, 8: intrinsic laryngeal muscles, 9: paralaryngeal muscles), salivary glands (10: glandular tissue, 11: margins of normal glands, 12: paraglandular fat spaces, 13: mandible and associated muscles), and lymphatic tissue (14: regional lymph node areas of the pharynx, 15: regional lymph node areas of the larynx, 16: regional lymph node areas of the salivary glands). Overall image noise, spatial resolution, and diagnostic acceptability were also evaluated.

In compliance with the guidelines mentioned above, each reader assessed the visually sharp reproduction of anatomical substructures on a two-point Likert-type scale (1: no, 2: yes), the image noise and spatial resolution on a three-point Likert-type scale (1: too little, 2: optimum, 3: too much), and the overall diagnostic acceptability on a four-point Likert-type scale (1: fully confident for diagnostic interpretation, 2: probably confident for interpretation, 3: confident only under limited conditions for visualization of abnormalities, 4: unacceptable). Image artifacts were also graded on a four-point Likert-type scale (1: no artifacts, 2: minor artifacts not affecting the visualization of any structure, 3: major artifacts affecting the visualization of normal structures, 4: artifacts affecting diagnostic information). Beam hardening and photon starvation artifacts caused by metal dental hardware were not evaluated. Inter-rater agreement was calculated based on the results obtained for each category and each acquisition technique.

The radiation exposure was quantified as pitch-corrected computed tomography dose index (CTDI_vol) and dose length product (DLP) as provided by the scanner. The effective radiation dose (ED) associated with the CT examination was calculated using the specific conversion factor for neck examinations in adults (0.0051 mSv × mGy⁻¹ × cm⁻¹) [11]. The estimated radiation exposure in the SECT group was matched to that of the DECT group in an ex ante trial using a dedicated 16-cm acrylic CTDI phantom by adjusting the reference tube current time product stepwise. Single-energy examinations were referenced to a 32-cm phantom; thus, values had to be converted to match the 16-cm phantom [11]. According to the vendor (Siemens Healthcare GmbH), the respective conversion factor for the CT system used in this study is 2.0 for 120 kV, but it must be adapted according to the chosen kilovolt value by TVA to a maximum of 2.4 at 70 kV.

Interval-level data were evaluated for normal distribution using the Shapiro–Wilk test. If the data were assumed to be normally distributed, values are given as mean ± standard deviation; otherwise, and in cases of ordinal-level data, values are given as median and interquartile range. For comparison of objective image quality data and radiation exposure between the SECT and DECT groups, a non-parametric Mann–Whitney U test was performed, as normal distribution could not be assumed according to the Shapiro–Wilk test. Comparison of subjective image quality ratings was performed using a Wilcoxon rank sum test. Inter-rater agreement was evaluated by calculating Cohen’s kappa value (κ); κ was interpreted according to Landis and Koch [12]. Results were accepted as statistically significant for p values < 0.05. The Statistical Package for the Social Sciences (SPSS) was used for randomization process and data analysis (IBM SPSS for Windows, Version 24.0 released 2016, IBM Corp.).

In SECT and DECT, ROIs for attenuation analysis could be drawn in all study participants (Fig. 2). Mean ROI size was 0.47 ± 0.18 cm² for vessel attenuation, 0.60 ± 0.13 cm² for muscle attenuation, and 0.33 ± 0.15 cm² for parapharyngeal fatty tissue. For DECT, a significantly increased CNR and CNRD was found for jugular veins relative to fatty tissue and for muscle tissue relative to fatty tissue compared to SECT (all p < 0.001). No significant difference was found for vessel attenuation relative to muscle tissue (CNR p = 0.36; CNRD p = 0.35). Table 3 summarizes all results from the objective image quality assessment.

All anatomical substructures evaluated were visually sharp and definable in all SECT and DECT examinations.

Image noise and spatial resolution were comparable between DECT and SECT for both readers. In contrast, overall diagnostic acceptability and image artifacts differed significantly between the DECT and SECT groups in favor of DECT for both readers. See Table 4 for a detailed overview and specific p values. Figure 3 provides a sample comparison of artifacts at the height of the shoulders between SECT and DECT.

Inter-rater agreement for image noise evaluation, overall diagnostic acceptability, and image artifacts was substantial for SECT and DECT (κ > 0.7). A moderate inter-rater agreement was found for spatial resolution (SECT and DECT κ = 0.6).

In the SECT group, TVA automatically selected 70 kV in 3 of the 40 patients (7.5%), 80 kV in 10 patients (25%), 90 kV in 19 patients (47.5%), and 100 kV in 8 patients (20%). Neither 110 nor 120 kV was selected. CTDI_vol, DLP, and ED were comparable between the study groups. Table 5 provides a detailed overview.