Comparability of size measurements of the pancreas in magnetic resonance imaging and transabdominal ultrasound

Transabdominal ultrasound (US) and magnetic resonance imaging (MRI) are commonly used for the examination of the pancreas in clinical routine. We therefore were interested in the concordance of these two imaging methods for the size measurement of the pancreas and how age, gender, and body mass index (BMI) affect the organ size.

Abdominal ultrasound (US) is an almost ubiquitously available and noninvasive method for investigation of abdominal organs like the pancreas. Therefore, it is often used as the first imaging method even when the patient has to undergo further imaging examinations. However, US is a highly operator-dependent technique, and imaging quality can be limited for example due to intestinal gas or marked obesity (Brown, Sirlin, Hoyt, & Casola, 2003). Other noninvasive imaging techniques for evaluation of the pancreas include computed tomography (CT) and magnetic resonance imaging (MRI). They produce cross-sectional images usually in transversal orientation. Although operator effects have been demonstrated for these techniques as well, they are less pronounced than in US, so that MRI measurements are often considered to be more reliable (Ryan, Semelka, Molina, Yonkers, & Vaidean, 2010;van Vliet et al., 2008).
Compared to CT, MRI provides better soft tissue contrast and in combination with magnetic resonance cholangiopancreatography (MRCP) it can be used for evaluating of the pancreatic duct (Bulow et al., 2014;Mensel et al., 2014). Validity of different imaging modalities has been assessed for cystic (Lee et al., 2015;Maimone et al., 2010) and solid lesions (Tummala, Junaidi, & Agarwal, 2011;Vukobrat-Bijedic et al., 2014) of the pancreas but there is still a lack of studies comparing the morphology and size of the anatomic segments of the pancreas, particularly for US. Due to advances in US technology and improved imaging quality this technique becomes more relevant for diagnostic and therapeutic strategies (Engjom et al., 2017;Lerch et al., 1992;Poza-Cordon & Ripolles-Gonzalez, 2014;Sun et al., 2017). Therefore, a comparison with high resolution imaging methods such as CT or MRI with US is justified.
We have used the Study of Health in Pomerania (SHIP), a population-based study, launched in 1997 in the region of West Pomerania in Northeast Germany, that follows two main objectives: first, it assesses prevalence and distribution of common risk factors and a broad range of subclinical disorders and clinical diseases and second, it analyzes their underlying associations (Ludemann et al., 2000). SHIP collects clinical and anthropometric data and is therefore suitable for analyses in the general population.
Our study has two objectives, that is, (a) to investigate the agreement between the two measuring methods for pancreatic size (US and MRI) and (b) to assess the effect of age, gender and body mass index (BMI) on pancreatic size measured by MRI. A better understanding of confounding factors and sources of bias will assist the interpretation of pancreatic size measurements in clinical settings.

| Study population
Participants were recruited from the SHIP-2 project, a second examination follow-up of the first SHIP cohort that was conducted between 2008 and 2012 (Volzke et al., 2011). In total 2,333 participants were recruited for SHIP-2, of which 1,182 (50.7%) were examined by MRI and 744 (31.9%) by US. None of the participants had any known previous pancreatic disorders and probands with a history of pancreatitis, anatomical variants such as pancreas divisum or pancreatic tumors were excluded. We only included subjects investigated by both MRI and US on the same day. After approval by the local institutional review  Table 1). The process of selection of individuals for the study was summarized by a flow chart (Supporting Information Figure S1).

| Transabdominal US
Measurements were carried out by 12 physicians, who were experienced examiners for ultrasonography and had undergone a training instruction to standardize the examination procedure for SHIP. This included an experience of at least US 2,500 examinations of the abdomen. US examination was performed using an ALOKA ProSound SSD-5000 SV ultrasound machine (Hitachi Medical Systems, Wiesbaden, Germany). Participants were examined in supine position after an overnight fasting period. The pancreas was accessed by subcostal view, the pancreatic head and body were measured in ventral-dorsal direction and the tail was measured perpendicular to the main axis of the organ (Supporting Information Figure S2A-C). Data were recorded electronically via a computer-based data entry mask. A cohort of 50 subjects was examined by two US examiners who separately measured all three anatomic segments of the pancreas to calculate interobserver variability for this method.  Scatterplots stratified by part of the pancreas (head, body, tail) were produced to visualize the association between US and MRI.

| Statistical analysis
Boxplots were created to investigate the distribution of US measurements per examiner. Bland and Altman plots were used to assess agreement between US and MRI measurements (Bland & Altman, 1986;Giavarina, 2015), whereby the mean of the two measurements (US and MRI) was plotted versus the difference. No adjustment was made for confounding variables gender, age, and BMI as both measurements were taken on the same subject (matched analysis).
The mixed command was used to adjust for the random effect of US examiner using a random-intercept model. Variance introduced by the examiner (reffects option of the pred command) was derived from this model and subtracted from the US measurements to remove the examiner effect. These adjusted US measurements were used to compute Bland-Altman plots again to assess the change compared to the estimates obtained for unadjusted measurements. The intraclass correlation coefficient (ICC) was derived from the model using the estat icc command and interpreted according to the same criteria as outlined above.
Linear regression (regress command) was used to predict MRI based pancreatic size of head, body and tail depending on the variables BMI, age and gender. Interactions were tested for all significant variables of the main effect model. A significance threshold of 0.05 was applied.
Margins plots were used to visually display the three possible two-way interactions between BMI, age and gender for each of pancreatic head, body and tail with fixed values of the third variable not included in the interaction (BMI and age: median; gender: males) (marginsplot command). Model fit was assessed by visual inspection of residuals.

| Concordance of MRI and US
Pancreatic size in each anatomical section was smaller when measured by US than by MRI (Table 2). The mean difference between MRI and US measurements was 0.39 cm for the head, 0.18 cm for the body, and 0.54 cm for the tail, which corresponds to a relative difference of 14.4-43.3% (Figure 1). The discrepancy was highest for the tail and lowest for the body. The wide 95% confidence intervals (CIs) further indicate a high level of inconsistency. Furthermore, Bland and Altman plots indicated certain trends in the data: First, the pancreatic head tended to show smaller measurements in US compared to MRI for smaller means of these two measurements. Second, for the body and tail, dispersion of the differences tended to be stronger for larger mean values of the two measurements.
US measurements were adjusted for the effect of examiner, given that considerable variability was observed between examiners in the boxplots (Figure 2 For the pancreatic head, size slightly increased with higher BMI but was not significantly different among different age groups of the study population ( Figure 4a) or between males and females ( Figure 4b). Size decreased with age in females but not in males (p = .03, Figure 4c). In contrast, a stronger increase of pancreatic body size with BMI was observed in males than in females, which was highly significant (p = .007) leading to lines drifting apart with higher BMI. An interaction of pancreatic body and BMI for different age groups or gender was not significant (Figure 5a,b). Pancreatic body size slightly decreased with higher age, but did not significantly differ between both genders ( Figure 5c). For the pancreatic tail, there was an increase of size with higher BMI, but no significant interactions were observed among different age groups (Figure 6a) or males and females (Figure 6b). Pancreatic F I G U R E 5 Margins plot for the interactive effect of the size of the pancreatic body measured by MRI (n = 330): BMI × age (a), BMI × gender (b), and age × gender (c). p-Values are derived from the respective interaction term F I G U R E 6 Margins plot for the investigated interactive effect of the size of the pancreatic tail measured by MRI (n = 330): BMI × age (a), BMI × gender (b), and age × gender (c). p-Values are derived from the respective interaction term tail size decreased with age in both genders which was more pronounced in females than in males (Figure 6c).

| DISCUSSION
Transabdominal US is an important diagnostic tool for a variety of gastroenterological disorders since it is usually used as the first diagnostic examination. It is ubiquitously available and easy to handle despite its dependence on the examiner's experience. In addition, quality of imaging in US has dramatically improved so that one can assume that this imaging method will gain more importance for diagnosis and interventional procedures in the future. Not all patients with pancreatic disorders automatically receive cross-sectional-imaging such as MRI, which usually is reserved for more complex questions. For example, patients with mild acute pancreatitis or for diagnostic workup for diabetes mellitus normally undergo US. That is why decision-making process of clinicians is often based on US as only imaging modality.
Therefore, we were interested in the accuracy of US measurements compared to a method with an acknowledged high imaging quality of the pancreas which is MRI (Bulow et al., 2014;Kuhn et al., 2015). By using the population-based SHIP, we recruited a large number of healthy individuals who were examined by both methods on the same day (Volzke et al., 2011).
There was a discrepancy between measurements by MRI and transabdominal US. Generally, US measured smaller sizes for each segment of the pancreas when compared to MRI. Differences ranged between 14.4 and 43.3% depending on the part of the organ and was greatest for the tail. Our study further demonstrated, that in 7% of the volunteers the pancreatic tail was not visible by US. To the best of our knowledge there are no comparative studies investigating the normal pancreas by both MRI and US. Organ size was assessed in previous studies using transabdominal US alone to define reference values for this technique (Sienz et al., 2011). In case of MRI, morphometry of the pancreas is usually performed by calculation of the pancreatic volume (Macauley et al., 2015;Virostko, Hilmes, Eitel, Moore, & Powers, 2016) and only one MRI study directly measured pancreatic anteriorposterior diameters (Sato et al., 2012). Second, our data showed a great variability of size measurements in US. One reason is that 12 different sonographers investigated the pancreas and we found a strong interobserver variability with only fair to poor agreement although all our imagers were trained for gastrointestinal US. In a study focusing on the quality of ultrasonography of adnexal masses performed by radiologists, comparably poor findings were obtained in spite of differences of their US experience and numbers of examinations. However, examiners who underwent subspecialization in female reproductive tract performed better (Levine et al., 2008). Unfortunately, studies on the quality of US performance for organs of the upper abdomen including the pancreas are lacking so that further investigations will be helpful.
The visual plane of the pancreas on ultrasonography may also impact measurements. Some authors have used other approaches such as a sagittal (Niederau et al., 1983) or cranio-caudal view (Pochhammer, Szekessy, Frentzel-Beyme, & Hollstein, 1984) or even an inclined position originating from a transversal view (Kolmannskog, Vatn, Swensen, Aakhus, & Gjone, 1983)  Our results further indicate that the size of the pancreas is dependent on anthropometric data, age, and gender of individuals. There are conflicting data regarding dependence of organ size on age (Sienz et al., 2011). No relevant alteration of size was observed in individuals with a healthy pancreas or with diabetes mellitus (Silva et al., 1993;Vossen, 1986), but the echogenity of the parenchyma was found to increase with age (Zimmermann, Frank, Weiss-Simon, Burkhard, & Seidel, 1981). Others reported an increase (Niederau et al., 1983) or a decrease of organ size (Ker, Tabata, Sheen, Jan, & Chen, 1984) with age. Measurements were mostly performed using sagittal diameter.
Our data indicate a decrease of organ size with increasing age, which is dependent on the gender of the volunteer. In females, the size of the pancreatic head and tail decreased more prominently with age compared to males. These data indicate that age effects are more pronounced in the pancreatic head and tail than in the pancreatic body.
In addition, there was an increase of pancreatic body size along with increasing BMI, which was more prominent in males than in females.
Our observations are in line with other studies that also reported a positive correlation between pancreatic size and BMI (Niederau et al., 1983). Irrespective of the organ segment, men showed larger sizes than women being comparable to observations from Niederau and co-workers (Niederau et al., 1983) while others have found a smaller pancreatic head in men (Silva et al., 1993). Possibly, the larger pancreas size in obese individuals results from fatty infiltration of the pancreas, analogous to development of fatty liver, caused by a progressive betacell dysfunction in the endocrine pancreas (Misra et al., 2015). Moreover, a positive correlation of pancreatic fat content with BMI has been described in the SHIP-TREND cohort, which was recruited between 2008 and 2015 in the same region like our study (Kuhn et al., 2015).
Since the three investigated predictors alone explained up to 34% of the variation in pancreatic size, it is hypothesized that reference values taking these individual factors into account may be useful in a clinical setting to assess pancreatic pathology. Apart from laboratory parameters that help for the selection of individuals and estimation of the number of study participants in clinical studies (Farkas et al., 2019) variations of anatomic sizes of the pancreas should also be considered. Indeed, in the context of the growing importance of imaging modalities this comparative study indicates that diagnostics based on US measurements should be followed by a confirmatory diagnostic procedure due to the risk of measurement bias.

| CONCLUSIONS
We conclude that measurements of the pancreas size performed by US need to be interpreted with caution as they differ to MRI, a crosssectional imaging method, which is considered to be more objective.
Second, there is a substantial variation of measurement results among ultrasonographers limiting the comparability of organ size evaluations.
Sources of individual variability due age, gender, and BMI should be considered when evaluating morphologic features to diagnose pathologic conditions. ACKNOWLEDGMENT