Development of size-specific institutional diagnostic reference levels for computed tomography protocols in neck imaging

Purpose: To develop size-specific institutional diagnostic reference levels (DRLs) for computed tomography (CT) protocols used in neck CT imaging (cervical spine CT, cervical CT angiography (CTA) and cervical staging CT) and to compare institutional to national DRLs. Materials and methods: Cervical CT examinations (spine, n = 609; CTA, n = 505 and staging CT, n = 184) performed between 01/2016 and 06/2017 were included in this retrospective study. For each region and examination, the volumetric CT dose index (CTDIvol) and dose-length product (DLP) were determined and binned into size bins according to patient water-equivalent diameter (dw). Linear regression analysis was performed to calculate size-specific institutional DRLs for CTDIvol and DLP, applying the 75th percentile as the upper limit for institutional DRLs. The mean institutional CTDIvol and DLP were compared to national DRLs (CTDIvol 20 mGy for cervical spine CT (DLP 300 mGycm) and cervical CTA (DLP 600 mGycm), and CTDIvol 15 mGy for cervical staging CT (DLP 330 mGycm)). Results: The mean CTDIvol and DLP (±standard deviation) were 15.2 ± 4.1 mGy and 181.5 ± 88.3 mGycm for cervical spine CT; 8.1 ± 4.3 mGy and 280.2 ± 164.3 mGycm for cervical CTA; 8.6 ± 1.9 mGy and 162.8 ± 85.0 mGycm for cervical staging CT. For all CT protocols, there was a linear increase in CTDIvol and DLP with increasing dw. For the CTDIvol, size-specific institutional DRLs increased with dw from 14 to 29 mGy for cervical spine CT, from 5 to 17 mGy for cervical CTA and from 8 to 13 mGy for cervical staging CT. For the DLP, size-specific institutional DRLs increased with dw from 130 to 510 mGycm for cervical spine CT, from 140 to 640 mGycm for cervical CTA and from 140 to 320 mGycm for cervical staging CT. Institutional DRLs were lower than national DRLs by 81% and 67% for cervical spine CT (dw = 17.8 cm), 43% and 51% for cervical CTA (dw = 19.5 cm) and 59% and 53% for cervical staging CT (dw = 18.8 cm) for CTDIvol and DLP, respectively. Conclusion: Size-specific institutional DRLs were generated for neck CT examinations. The mean institutional CTDIvol and DLP values were well below national DRLs.


Introduction
The volumetric computed tomography dose index (CTDI vol ) and dose-length product (DLP) of a computed tomography (CT) examination can be used as guidance to review the radiation application. National and institutional diagnostic reference levels (DRLs), which are used for quality assurance in radiology departments, are most commonly based on CTDI vol and DLP [1][2][3][4]. In the German radiation protection ordinance, DRLs are defined as dose indices for typical examinations employing ionising radiation, based on standardised phantoms or groups of patients with standard morphometry, while using a suitable modality and procedure for a respective examination [5]. DRLs are established as the 75th quartile of a distribution of patient doses for different users and CT scanner vendors; however, they are not ideal dose indices. Although DRLs are not fixed radiation limits, their constant and unjustified exceedance is not allowed and requires disclosure. In particular, due to the renewed German radiation protection ordinance and act, the dose indices of individual patients and of patient groups now need to be constantly evaluated and reviewed with respect to published DRLs to identify dose incidents and poor examination protocols-even if patients do not match the weight of a standardised patient (70±3 kg), and even though DRLs should not be used on an individual patient basis [6,7].
Most current national DRLs (nDRLs) have major drawbacks in their application and significance [8]. First, the CTDI vol is based on radiation exposure measurements by means of a standardised cylindrical CTDI-phantom (either with a diameter of 16 cm, representing a head, or with a diameter of 32 cm, representing a body). The cylindrical shape and the material (poly-methyl-meth-acrylate) are only a rough approximation of the human body. Second, DRLs for adults are typically provided for average-sized patients, although it is well known that the effective dose of a CT examination depends on patient characteristics, such as the body weight and height, body mass index (BMI) and patient composition [9,10]. High dose values in small patients might go unnoticed as CTDI vol or DLP values do not exceed national DRL thresholds. Particularly for these smaller patients, size-specific DRLs are needed. Third, national DRLs are sometimes only provided for anatomic regions rather than for dedicated CT protocols, although the dose may vary notably within one anatomic region [11]. For neck CT imaging, three different protocol-specific DRLs are provided. However, for abdominal imaging, the same DRL holds for CT examinations to diagnose a suspected kidney tumour or metastasis versus a suspected kidney stone; although the latter indication can be answered with a considerably reduced radiation exposure. The latter is reflected in the EUROSAFE imaging campaigns' desire to implement CT protocol-specific DRLs [12]. Recently, size-dependent DRLs based on the BMI or patient diameter measurements were proposed for body regions such as the chest or abdomen [8,13,14]. However, dose variation among different CT protocols and patient sizes in neck CT imaging remains unknown, and dedicated institutional DRLs (iDRLs) may improve CT quality assurance.
Therefore, the aim of our study was to analyse CT dose data from different neck CT imaging protocols to generate size-specific and CT protocol-specific institutional DRLs.

Materials and methods
This retrospective study was approved by the institutional review board. All CT examinations assigned to the national DRLs for CT of the neck from the four institutional multi-detector CT scanners (Definition Flash with 128 slices, Definition AS+ with 128 slices, Definition AS with sliding gantry with 64 slices, Definition Edge with 128 slices, Siemens Healthineers, Forchheim, Germany) performed between 01/2016 and 07/2017 were included in our study (cervical spine CT, cervical CT angiography (CTA) and cervical staging CT). The scan volumes of the institutional standard CT protocols were as follows: external auditory canal to lower edge of the clavicle for cervical staging CT; cranio-cervical transition to the first thoracic vertebra (T1) for cervical spine CT; and aortic arch to crown for cervical CTA. Cervical spine CT examinations were performed without contrast enhancement (iodine), whereas contrast agents were applied during cervical staging CT (venous phase) and cervical CTA (arterial phase). Common indications for cervical spine CT examinations were trauma, radiculopathy, myelopathy or osseous metastases. Indications for cervical CTA were stroke, suspected vessel stenosis, occlusion or aneurysm. Indications for cervical staging CTs were suspected tumours in the neck region, lymph node staging or abscesses.
For the chosen CT protocols, there were 8328 examinations performed in the study interval. The exclusion parameters for this study were paediatric patients, patients with a water-equivalent diameter (d w )<15 cm or d w 25 cm, or reconstructed images with truncated field of view. After exclusion, 609 cervical spine CT examinations, 184 cervical staging CT examinations and 505 cervical CTA examinations were included in our analysis (figure 1).
CT dose monitoring and calculation of d w A dose management system (DoseIntelligence, Pulmokard, Herdecke, Germany) based on the Digital Imaging and Communications Radiation Dose Structured Report (DICOM-RDSR) was used to obtain CT dose data and technical parameters, including CTDI vol , DLP, pitch, mean tube current-time product, tube voltage, scan length, anatomic region, CTDI-phantom size and the CT protocol [4]. Patient height and weight were automatically stored in the DICOM-RDSR at the time of the CT examination if they were available in the electronic patient records (only in 5% of the examinations). Assignment of CT protocols to body regions (which was necessary for comparison to national DRLs) was performed manually by one radiologist (J.B. with five years of experience in radiology).
The method of d w calculation by means of a previously validated self-designed algorithm and the Matlab environment (version R2015, The Mathworks, Natick, MA) was previously published [8]. In short, the algorithm first removes the CT scanner table, then identifies the outer body circumference, and finally, calculates the d w according to the Americam Association of Physicists in Medicine (AAPM) guidelines [8,15]. Truncated images were identified by analysis of pixels other than air adjacent to the outer border of the field of view [8].
Data sets with an average number of surrounding pixels other than air >100 were excluded to ensure that truncation of CT images did not influence our results [8]. For the purpose of this study, the mean d w of the scan volume was used. Subsequently, the calculated d w was stored into the institutional CT dose management system (DoseIntelligence, Pulmokard, Herdecke, Germany).

Image acquisition and reconstruction
CT examinations were performed in spiral mode without or with intravenous contrast material (Omnipaque, GE Healthcare, Munich, Germany) depending on the indication. Automated tube current modulation (CareDose 4D) was activated in all examinations, whereas automated tube voltage selection (CarekV, both Siemens Healthineers, Forchheim, Germany) was activated in most examinations. Images were reconstructed with a medium level of iterative reconstruction (SAFIRE™ Level 3, Siemens Healthineers, Forchheim Germany) when performed on the Somatom Definition AS+ (scanner 1), Flash (scanner 2), or Edge (scanner 3) CT scanners, and with filtered back-projection on the Somatom Definition AS (sliding gantry, scanner 4) CT scanner. CTDI vol values obtained from the dose protocol were based on the 32 cm phantom for all examinations in this study [16].

Calculation of iDRLs and data analysis
For calculation of the iDRLs, patients were categorised into 1 cm d w bins (e.g. bin d w =15 cm includes all d w in between 15 cm and 16 cm). Analysis was performed per neck CT protocol separately. Linear regression was used to calculate iDRLs based on patient d w .
As recommended by the International Commission on Radiological Protection (ICRP), the 75th percentile of the CTDI vol and DLP were regarded as the upper limit of the iDRLs, and the 25th percentile was chosen as the lower limit [17].
In addition to calculation of iDRLs, mean institutional CTDI vol and DLP were calculated for an average-sized patient as determined by the d w for each neck CT protocol from the corresponding linear regression curves and were compared to the nDRLs. The average-sized patient by means of the d w was calculated by calculation of the mean d w over all patients for each of the three CT protocols (table 1). Subsequently, mean institutional CTDI vol and DLP values for the calculated d w were determined from the linear fits through the 1 cm d w bins described in the last section (solid lines in figures 2-4).
Additionally, iDRLs of both the CTDI vol and DLP were compared to the size-specific DRLs of the American College of Radiology Dose Index Registry (ACR-DIR), which are provided for different d w groups [19]. For cervical spine CT, they range between 24 and 28 mGy for the CTDI vol , and between 495 and 575 mGycm for the DLP. For cervical staging CT size-specific DRLs range between 18 and 19 mGy (CTDI vol ) and 509 to 560 mGycm (DLP). Furthermore, the number of CT examinations exceeding the national DRLs and the size-specific iDRL were compared.

Differences between the CT scanners and data analysis
Impact of the employed CT scanner on the CTDI vol of each CT protocol was analyzed. All data is given as the mean±standard deviation with range, or median values. Linear regression was performed to generate iDRLs. The institutional mean CTDI vol and DLP are provided with one decimal precision, whereas iDRLs are rounded to the nearest integer. Statistical analysis for comparison of the DLP, CTDI vol , d w and scan range was performed using a student's t-test with a significance level of p<0.05.
Weight information was only available in the electronic report in 5% of the examinations. For these patients, the mean patient weight was 74±14 kg.  were significant between the three protocols (p=0.4999 for the comparison of the CTDI vol between cervical CTA and cervical staging CT, else p=0.05).

Relationship between CTDI vol , DLP and d w
For all CT protocols, there was a linear increase in CTDI vol and DLP with increasing d w (see     The 50th and 75th quartile of the CTDI vol and DLP published by the ACR-DIR for staging CT (neck with contrast material) and cervical spine CT (without contrast material) were compared to the institutional 50th and 75th quartile of the CTDI vol and DLP (table 4) [19].
For cervical staging CT, the institutional 50th and 75th quartiles were 27%-53% lower than the ACR-DIR DRLs for CTDI vol . The differences were even larger for the DLP, ranging between 41% and 70%. With increasing d w , differences between dose levels of the current study and of the ACR-DIR decreased. For cervical spine CT, the institutional 50th and 75th quartiles for the CTDI vol were 20%-36% smaller for the d w groups between 13 and 21 cm compared to ACR-DIR dose levels, whereas for the largest d w group, the 50th and 75th quartiles were similar to ACR-DIR dose levels. The institutional 50th and 75th quartiles for the DLP differed considerably throughout all the d w groups: they were 32%-68% lower than the ACR-DIR dose levels. Again, differences decreased with increasing d w .

Outlier analysis
There were 286 (251) examinations with CTDI vol (DLP) exceeding the size-specific iDRLs, whereas there were only 75 (74) examinations exceeding the nDRLs (see table 5 for the individual CT regions and figure 5 for cervical spine CTs). Patients with CTDI vol or DLP>iDRLs but not reaching the nDRLs had an average d w lower than the average d w of patients with CTDI vol and DLP<iDRLs. In return, patients with CTDI vol or DLP>nDRLs had an average d w higher than the average d w of the patient group with CTDI vol and DLP<iDRLs (see table 5).

Differences between the CT scanners
On average, the highest average CTDI vol and DLP values were obtained on scanner 4, except for cervical staging CT examinations. Please note that only filtered back-projection was available on scanner 4. For cervical staging CT, the CTDI vol was highest for scanner 2. Here, all examinations were performed at 120 kV p (no automatic tube potential selection), whereas for all other scanners the tube potential varied between 80kV p and 120kV p for cervical staging CT examinations.

Discussion
In this study, a large-scale analysis of neck CT examinations was performed which included CT dose parameters and patient d w and consequently allowed for the development of sizespecific and CT protocol-specific iDRLs. Currently, there are three different national German DRLs provided for neck CT examinations, which are related to patient groups of standardised morphometry (70±3 kg). In our study, a strong linear relationship between the d w and CTDI vol and between the d w and DLP was found in all evaluated CT protocols: with increasing d w , both the CTDI vol and DLP values increased linearly. The linear relationship is an expected result, since the tube current modulation depends on the density of the patient as obtained by the scout view (topogram). The graphs demonstrate that the automatic exposure control is working and reduces the exposure for small patients for an individual CT protocol. In our institution, cervical spine CT examinations resulted in the highest CTDI vol , although the smallest d w were determined from these examinations. Although all protocols cover the neck region, the specific scan range is protocoldependent, leading to differences in d w : cervical CTA covers a larger range of the shoulders, causing a high d w , whereas cervical spine CT covers the smallest range of the shoulders (see the protocol description in the materials and methods section), leading to the lowest d w . Furthermore, the scan protocol parameters of cervical spine CT included the highest reference tube current-time product compared to the other two protocols. Due to the long scan coverage, the largest average DLP was found for cervical CTA. The average CTDI vol and DLP for cervical staging and CTA acquisitions were only 40%-60% of the nDRLs, whereas the CTDI vol and DLP were 70%-80% of the nDRLs for cervical spine acquisitions. Since the CTDI vol of cervical staging CT and cervical CTA are well below the nDRLs, these also cause low DLP values since the scan range is defined in the standard operating procedure and is fairly equal for all examinations within one protocol.
Compared to the DLP and CTDI vol values published for the cervical staging and spine CT examinations by Kanal et al from the ACR-DIR, our calculated 50th and 70th quartiles were considerably lower (−20% to −70%) for all d w bins, except for CTDI vol values in patients with a very large d w for cervical spine CT [19]. Here, the 50th and 70th quartiles were similar.
Our presented study and the performed analysis demonstrate the ability to perform a detailed subgroup analysis when using size-specific DRLs based on d w . Of note, Kanal et al based their calculations of d w on the localiser images (topograms). Calculations of d w based on the localiser image may slightly overestimate d w (by 4%) compared to calculations based on CT images, which is regarded as the reference standard and was performed in our study [20]. The 4% overestimation due to the calculation of d w cannot however explain the differences between the ACR-DIR data and our iDRLs (−20 to −70%). They are most likely due to differences in either scan protocol settings (used tube potential or tube current-time product) or chosen scan coverage. Unfortunately, the paper from Kanal et al does not provide this information [19].
As was observed in the outlier analysis, the largest percentage (12%) of examinations exceeding the national and institutional DRLs was found for cervical spine CT. Here, only acquisitions on scanner 4 resulted in CTDI vol values exceeding the nDRLs, whereas all acquisitions performed on scanners 1, 2 and 3 were below the nDRLs. The scan length for cervical spine CT for all d w amounts to only 12-15 cm. Hence, the increased DLP values are the result of an increased CTDI vol , but are not due to a too large scan coverage. The increased CTDI vol on scanner 4 is most likely the result of the missing availability of iterative reconstruction. Iterative reconstruction allows for reduction of the radiation exposure by approximately 20%-40% while maintaining the same image quality [21][22][23][24]. Although the average CTDI vol was well below the nDRLs for cervical staging CT scans on scanner 2, we noticed a missing automatic tube potential selection, which resulted in a higher CTDI vol compared to the other scanners. Consequently, the CT protocol was modified on scanner 2 to allow for automatic tube potential.
Depending on the CT protocol, 75%-85% of all included cervical CT examinations were below nDRLs, 6%-26% of which exceeded the size-specific iDRLs, depending on the CT protocol. This comparison also includes patients, which do not match the weight of the standardised patient with 70±3 kg. Our analysis demonstrated that a high d w causes the CTDI vol and DLP to be larger than institutional and national DRLs. For intermediate d w, the CTDI vol and DLP are often lower than institutional and national DRLs. The range for CTDI vol and DLP to be between institutional and national DRLs is small and includes only small d w (yellow dots in figures 4 and 5). These latter examinations would go unnoticed when using solely nDRLs for dose analysis. Hence, it is necessary to facilitate size-specific iDRLs, especially for patients smaller than the standardised patient is. Our results indicate that sizeand protocol-specific DRLs may help to improve CT quality assurance by allowing for a more detailed analysis. The nDRLs are useful to detect systematic dose application errors but do not allow for a comprehensive analysis. Inclusion of the patients' d w in the evaluation of CT radiation exposure increases the accuracy and reliability of institutional quality assurance, since size-specific DRLs allow us to detect dose outliers of small patients, whose dose values might still be below the nDRLs, but far from being optimal. Of note, such detailed analysis may require additional time and effort, but can be facilitated by using (semi-)automatic dose management systems [25].
In recent publications, iDRLs have been calculated by means of size-specific dose estimates (SSDEs) for several body regions [15,19,26]. SSDEs are calculated by multiplication of the CTDI vol and a patient size-specific conversion factor, which takes the patient size and attenuation into account [10,20,27]. Typically, a comparatively large d w results in a smaller SSDE than CTDI vol (conversion factor <1), whereas patients with a comparatively small d w result in an SSDE>CTDI vol . Very recently, the AAPM task group No. 293 has published a new report about SSDEs for head CT, following the report about SSDEs for body CT examinations [27,28]. However, CT protocols of the neck include parts of the head and the body, and are thus neither full head nor body CT examinations. Hence, the value of the d w varies extensively if calculated from the mean d w of the scan volume or from a single image. Until now, dedicated size-specific conversion factors for neck CT examinations with instructions on the calculation of the d w have not been published by the AAPM [20].
Our study has some limitations. First, it was necessary to exclude a relatively large number of examinations. In cases of truncated images, large patients might seem to have a smaller d w than their actual d w with a proportionately high CTDI vol . The exclusion of CT examinations resulted in a smaller number of patients. However, it seems unlikely that the exclusion of patients with an incomplete field of view may have caused a systematic error. Second, since only CT scanners from one vendor were available in our institution, the developed DRLs might be different for other vendors, since the automatic tube current modulation varies per vendor [29][30][31][32]. Third, weight data of patients was not available for all patients in the electronic patient records (only in 5% of the patients). As a result, patient weight could not be extensively evaluated and associated with the exposure data obtained for the different protocols. This is unfortunate since the nDRLs are based on a patient weight of 70±3 kg, and comparison of patient weight would have been of great interest.

Conclusion
Size-specific iDRLs were generated for cervical CT examinations based on the CTDI vol and DLP. Mean institutional CTDI vol and DLP values were well below the nDRLs. The iDRLs allow for a comprehensive analysis of CT radiation exposure with regard to the specific CT cervical protocol and the patient size.