European Radiology https://doi.org/10.1007/s00330-018-5906-0 EMERGENCY RADIOLOGY Revised Atlanta classification for CT pancreatic and peripancreatic collections in the first month of acute pancreatitis: interobserver agreement Neesmah Badat 1 & Ingrid Millet 2,3 & Lucie Corno 1 & Wassef Khaled 1 & Isabelle Boulay-Coletta 1 & Marc Zins 1 Received: 12 August 2018 / Revised: 17 October 2018 / Accepted: 22 November 2018 # European Society of Radiology 2019 Abstract Purpose To assess interobserver agreement when using the revised Atlanta classification (RAC) to categorize pancreatic and peripancreatic collections during the first month of acute pancreatitis (AP), and to correlate type of collection to outcome. Material and methods This retrospective study of 115 consecutive patients admitted for 123 AP episodes, 178 CTs performed within the first month showed peripancreatic abnormalities. Each AP episode was classified as mild, moderately severe, or severe based on the RAC. Two radiologists, blinded to clinical data, used RAC criteria to retrospectively categorize the collections as acute peripancreatic fluid collections (APFC) or acute necrotic collections (ANC). Interobserver agreement was assessed based on Cohen’s κ statistics and compared according to CT timing. Results Interobserver agreement for categorizing peripancreatic collections was moderate (κ = 0.45) and did not improve with time to CT (κ values, 0.53 < day 3, 0.34 on days 3–6, and 0.43 ≥ day 7). For detecting parenchymal necrosis, interobserver agreement was also moderate (κ = 0.45). AP was less severe in patients with APFC versus ANC (p = 0.04). Conclusion Our finding of moderate interobserver agreement when using the RAC to categorize pancreatic and peripancreatic collections by CT indicates that the accurate diagnosis of APFC or ANC by CT in the first 4 weeks after symptom onset is often challenging. Key Points • Interobserver agreement was moderate for categorizing peripancreatic collections. • Interobserver agreement did not improve with time from onset to CT. • Interobserver agreement was moderate for detecting parenchymal necrosis. Keywords Acute necrotizing pancreatitis . Multidetector computed tomography . Interobserver variability . Outcomes assessment . Pancreatitis Abbreviations ANC Acute necrotic collection AP Acute pancreatitis APFC Acute peripancreatic fluid collection CT IEP RAC WON * Ingrid Millet i-millet@chu-montpellier.fr Introduction 1 Department of Radiology, Hôpital Saint-Joseph, 185 Rue Raymond Losserand, 75014 Paris, France 2 Department of Medical Imaging, CHU Lapeyronie, 371 Avenue du Doyen Gaston Giraud, 34295 Montpellier Cedex 5, France 3 Montpellier University, Montpellier, France Computed tomography Interstitial edematous pancreatitis Revised Atlanta classification Walled-off necrosis Acute pancreatitis (AP) is a common and potentially severe condition. Establishing the diagnosis and prognosis of AP relies particularly on imaging studies and, more specifically, on computed tomography (CT) [1]. The Atlanta classification issued in 1992 [2] used CT findings to classify the manifestations of AP, introducing Eur Radiol standardized terms to designate each and thereby facilitating communication among the various specialists involved in managing patients with AP. However, sources of confusion were identified and poor interobserver agreement reported [3]. Additional reasons for revising the classification were advances in imaging techniques and the introduction of new radiological, endoscopic, and surgical treatment methods [4–6]. The revised Atlanta classification (RAC) was issued in 2012 [7, 8]. It introduces new terminology to designate AP types and stages, collections that develop during AP, and degrees of severity. The RAC differentiates interstitial edematous pancreatitis (IEP) and necrotizing pancreatitis, which in turn is classified as parenchymal necrosis, peripancreatic necrosis, or both. Pancreatic and peripancreatic collections are named based on two criteria, namely, whether they develop within or after the first 4 weeks and whether they contain fluid only or also solid material [6]. Acute peripancreatic fluid collection (APFC) designates a collection containing fluid only that developed within the first 4 weeks. An acute necrotic collection (ANC) contains both fluid and necrotic material and develops within the first 4 weeks. Collections present after the first 4 weeks and having a well-defined wall are pseudocysts if they contain fluid only and a walled-off necrosis (WON) if they also contain solid material. APFCs and pseudocysts occur in patients with IEP, whereas ANCs and WON are manifestations of necrotizing pancreatitis. Differentiating fluid collections from collections containing necrotic material, most notably APFCs vs. ANCs during the first month after AP onset, is crucial to determine the type of pancreatitis, optimize outcome prediction, and make the best treatment decisions [9, 10]. The distinction is particularly important in the absence of parenchymal necrosis, as it determines whether the patient has IEP or peripancreatic necrosis alone, two conditions with markedly differing outcomes [11, 12]. During the first week of AP, separating APFCs from ANCs may be challenging [13], because both types of collection may present as more or less homogeneous hypoattenuating areas without contrast enhancement. A 2016 report indicated poor interobserver agreement when the RAC was used to define the type of AP, detect peripancreatic necrosis, and determine the features of collections [14]. In this study, agreement improved with time from symptom onset to CT scanning; however, time to CT was up to 3 months, and the radiologists were not blinded to this variable. The objective of this study was to assess interobserver agreement when using the RAC to categorize pancreatic and peripancreatic collections on CT scans of patients with AP of less than 4-week duration. We also evaluated potential correlations between type of collection and disease severity. Materials and methods This retrospective study was approved by our institutional review board, which waived the requirement for informed consent, in accordance with French legislation on retrospective studies of anonymized data. Patients Between January 2014 and December 2015, 352 CTs were performed in 195 consecutive patients older than 18 years and admitted to our hospital for AP. According to the RAC [7], we defined AP as the presence of at least two of the following three features: abdominal pain consistent with AP, serum lipase activity at least 3-fold higher than the upper limit of normal, and imaging findings typical for AP. Of the 352 CTs, 277 were obtained within the first 4 weeks of AP and were reviewed by a radiologist (IB, with 20 years of experience in abdominal CT), who excluded the 82 CTs without peripancreatic abnormalities, 3 CTs performed to assess autoimmune pancreatitis and 14 unenhanced CTs, leaving 178 CTs for the study. Figure 1 is the patient flow chart. The 178 CTs were obtained in 115 patients, including 6 with two AP episodes and 1 with three AP episodes, separated by at least 6 months, with a normal CT in the interval. The following data were collected from the medical records: sex, age, admission date, time of abdominal pain onset, serum lipase activity upon admission, hospital stay length, cause of AP, need for invasive intervention, organ failure, and in-hospital mortality. The cases with collections visible by CT were divided into three groups depending on whether time from symptom onset to CT: 0–3, 3–7, and > 7 days. Each AP episode was classified as mild, moderately severe, or severe based on criteria adapted from the RAC [7]. Patients with mild AP had no organ failure, local complications (e.g., symptomatic collections and venous thrombosis), or systemic complications. Moderately severe AP was defined as the presence of transient organ failure (including systemic inflammatory response syndrome) or of local or systemic complications without persistent organ failure (< 48 h). Persistent organ failure lasting for more than 48 h was the defining characteristic of severe AP. CT imaging The entire abdomen and pelvis was scanned, using a 64detector machine (Lightspeed VCT 64; GE Medical Systems) with 1.375 pitch, 0.7 tube rotation time, 120 kV, and 130 to 700 mA depending on body habitus. Patients without contraindications received an intravenous iodinated contrast agent (Iomeprol, Iomeron 300, Bracco Imaging; or Eur Radiol Fig. 1 Flow chart Iopromide, Ultravist 300, Bayer) in a dose of 1.5 mL/kg at a flow rate of 3 mL/s. Images were first acquired without contrast material and with a nominal section thickness of 0.625 mm, reconstruction section thickness of 1.25 mm, and 1.25-mm increment. Then, enhanced images were acquired at the pancreatic phase (45 s after injection) and portal phase (70 s after injection), with a nominal section thickness of 0.625 mm, reconstruction section thickness of 1.25 mm, and 1.25-mm increment. Oral contrast material was not used. Most CTs (n = 145, 81%) were triphasic, 23 (13%) were biphasic (unenhanced and portal phase), and 10 (6%) were acquired only in the portal phase. It is our practice to perform an unenhance phase in patients with AP to readily detect biliary stones or parenchymal pancreatic calcifications, and an arterial phase to look for pseudoaneurysms. The CT images were screened for the following signs: peripancreatic fat stranding without collection; peripancreatic collections (existence and number); location and size of the largest collection; content of the collections recorded as homogeneous or heterogeneous; visible collection wall with its features; most appropriate term for designating the collection according to the RAC (APFC or ANC); surface area of parenchymal necrosis (< 30%, 30–50%, or > 50% of total surface area); overall volume of the pancreas; pancreatic calcifications; dilation of the main pancreatic duct recorded as focal or diffuse; pancreatic tumor; and portal, splenic or mesenteric venous thrombosis. Interobserver disagreements about the presence of fat stranding or peripancreatic collection (n = 12, 12/178, 7%) and about the most appropriate term for the collection (n = 33, 33/133, 25%) were resolved by consensus developed with a third abdominal radiologist (MZ, 28 years of experience). Interpretation of CT images Statistics Two senior abdominal radiologists (IM, with 10 years of experience, and LC, with 9 years of experience, hereafter designated readers 1 and 2, respectively) independently reviewed each CT on a dedicated picture archiving and communication system unit (Carestream Vue, version 11.3; Carestream Health). Multiplanar reconstructions were performed using the axial, coronal, and sagittal planes. The radiologists knew that the patients had AP but were unaware of other patient data, including the degree of AP severity. SAS software version 9.3 (SAS Institute) was used for the statistical analyses. Continuous variables were described as mean and standard deviation if normally distributed and as median and interquartile range otherwise. Categorical variables were described as number and percentage. Interobserver agreement for each CT finding was assessed by computing the kappa coefficient for binary variables and the weighted kappa coefficient for multi-class variables between the 2 radiologists. Results were interpreted as follows: Eur Radiol κ = 0–0.20, slight agreement; κ = 0.21–0.40, fair agreement; κ = 0.41–0.60, moderate agreement; κ = 0.61–0.80, substantial agreement; and κ = 0.81–1.00, almost perfect agreement [15]. The agreement rate between the readers was also calculated for each screened CT sign. Agreement rate regarding the most appropriate term for the collection was assessed according to days from symptom onset to CT (0–2, 3–7, or > 7) by applying the chi-squared test. Consensus reading was not used to compute interobserver agreement but to determine the type of collection when the two readers disagreed. This determination was used for the statistical analysis of correlations between type of collection and clinical outcome. Associations between the most appropriate term for the collection on the first CT scan in each patient (n = 123 CTs) and AP severity were evaluated separately for each reader and for the consensus using the chi-squared test. McNemar’s exact test was applied to compare the distribution of AP severity classes for each type of collection, between each individual reader and the consensus. All statistical tests were two-sided and p values < 0.05 were considered to indicate significant differences. Results Patients We included 115 patients, 69 males and 46 females, who ranged in age from 20 to 92 years (mean age, 58.2 ± 17.3 years; median age, 60; interquartile range (IQR), 45– 70 years) and experienced a total of 123 AP episodes. The main causes of AP were gallstones (n = 39, 32%) and alcohol abuse (n = 39, 32%). The other causes were complications of endoscopic retrograde cholangiopancreatography (n = 14, 11%), pancreatic tumor (n = 4, 3%), surgery (n = 4, 3%), and miscellaneous conditions (hypertriglyceridemia, drugs, and sphincter Oddi dysfunction) (n = 9, 7%). No cause was identified in 14 (11%) patients. Median hospital stay length was 8 days (IQR 5–14 days). Mean time from pain onset to first CT was 3.2 days (median, 3 (2, 3) days). Mean serum lipase activity was 3243 IU/L (median 1491 (411–5116) IU/L). A single patient had infection of a peripancreatic collection. Six (4.5%) patients died during their hospitalization. Interobserver agreement Table 1 reports the κ values reflecting interobserver agreement for each CT finding. Agreement was substantial for presence of a collection (κ = 0.80). Moderate agreement was found for content of the largest collection (κ = 0.51), most appropriate term for the collection (κ = 0.45), and parenchymal necrosis (κ = 0.45). Table 2 shows the results obtained when interobserver agreement for the most appropriate term for the collection was assessed according to the time from pain onset to CT. The κ values ranged from 0.34 to 0.53 and the percentage of agreement from 65 to 87%. Agreement was lowest between days 3 and 7. Agreement did not vary significantly with time to CT (p = 0.07). Type of collection and clinical outcome Of the 123 episodes of AP, 6 (5%) did not have an assessment of AP severity, due to missing clinical data. Of the remaining 117 CT scans, 87 (75%) were obtained during mild AP, 25 (21%) during moderately severe AP, and 5 (4%) during severe AP. The consensual reading detected at least one parenchymal or peripancreatic collection for 97 (97/123; 79%) CTs, including 40 with APFCs and 57 with ANCs. Regarding the consensual reading, the proportion of mild AP episodes was 84% (32/38) when APFCs were seen and 62% (33/53) when ANCs were seen. Corresponding proportions for moderately severe AP were 11% (4/38) and 32% (17/53). These clinical outcome distributions translated into greater severity in patients with ANCs compared to APFCs (Fig. 2), with a nonsignificant trend for reader 2 (p = 0.09) and significant differences for reader 1 (p = 0.03) and the consensus (p = 0.04). No significant differences for clinical outcome distributions were found between the consensus and reader 1 or between the consensus and reader 2 (p = 0.46 and p = 0.71, respectively) (Figs. 3, 4, 5, and 6). Discussion Interobserver agreement was only moderate when the RAC was used to classify pancreatic and peripancreatic collections seen on CT scans during the first month of AP. No improvement in interobserver agreement occurred with increasing time from AP onset to CT. Finally, the type of collection seemed to be associated with AP severity. Our findings are consistent with a study comparing local radiologists and an expert radiologist, who used the RAC to read CTs obtained within 3 months of AP onset [14]. Interobserver agreement was only fair for peripancreatic necrosis (κ = 0.326), features of collections (κ = 0.408), and the most appropriate term for designating collections (κ = 0.356). In this study [14], agreement was better for CTs obtained at least 2 weeks after AP onset. In contrast, in our study, agreement did not improve with increasing time to CT. This apparent discrepancy may be related to the time frame difference, with CTs included only within the first 4 weeks in our study compared to 3 months in the earlier study. Within 4 weeks, only two types of collections can be found, APFCs and ANCs. Eur Radiol Table 1 Interobserver agreement for CT findings CT findings Reader 1 Reader 2 N (seen by both readers) Agreement (%) Kappa 95% CI Presence of a collection 141 137 133 166/178 (93.3) 0.80 (0.70–0.91) 86/133 (64.7) 0.70* (0.62–0.78) 105/133 (78.9) 0.51 (0.35–0.67) 111/133 (83.5) 0.64 (0.51–0.77) 30/36 (83.3) 0.63 (0.37–0.89) Number of collections 1 2 3 35 39 18 36 34 14 22 16 6 >3 49 53 42 Content of the largest collection Homogeneous Heterogeneous Presence of a wall 39 47 27 102 39 90 55 78 36 Wall margins Complete 10 14 9 Incomplete 26 22 21 100/133 (75.2) 0.45 (0.30–0.59) APFC ANC Parenchymal necrosis 55 78 28 26 107 41 24 76 19 147/178 (81.8) 0.45 (0.29–0.61) < 30% 30–50% > 50% Parenchymal calcifications$ Main pancreatic duct dilatation 22 4 2 29 18 39 2 0 26 35 20 17 161/176 (91.5) 159/178 (89.3) 0.68 0.59 (0.52–0.83) (0.42–0.75) 12/17 (70.6) 0.41 (0.01–0.84) 9 9 22 13 6 6 13 16 17 9 9 8 166/178 (93.3) 169/178 (95) 0.56 0.62 (0.34–0.78) (0.39–0.84) Most appropriate term for the collection Type of dilatation Focal Global Pancreatic tumor Portal venous thrombosis CT computed tomography, 95% CI 95% confidence interval *Weighted kappa $ 2 CTs did not include unenhanced acquisitions When examined after a longer interval, however, patients may have pseudocysts and/or WON, which may be easier to identify. Furthermore, the readers in the earlier study were not blinded to time to CT. However, our finding of worse agreement between days 3 and 7 is unexpected. One possible explanation is selection bias, with a higher proportion of patients with CT within the first 3 days developing ANC. Table 2 Interobserver agreement for the most appropriate term for designating a collection (APFC or ANC) according to time from pain onset to CT The low contrast resolution of CT raises challenges in detecting necrotic material within collections. Some measure of subjectivity is involved in assessing whether the content of a collection is homogeneous or heterogeneous as defined in the RAC. The CT findings defined in the RAC to differentiate APFC from ANC are insufficiently precise and do not permit an accurate characterization of peripancreatic collections. This Timing of CT Number of CTs Number of CTs showing collections Agreement (%) Kappa 95% CI 0–72 h Day 3–day 7 > day 7 Total 63 76 39 178 50 52 31 133 39/50 (78%) 34/52 (65%) 27/31 (87%) 0.53 0.34 0.43 [0.31–0.76] [0.15–0.52] [− 0.02–0.89] CT computed tomography, 95% CI 95% confidence interval Eur Radiol Fig. 2 Percentage of APFCs (a) and ANCs (b) depending on disease severity. R1: reader 1; R2: reader 2. APFC, acute pancreatic fluid collection; ANC, acute necrotic collection fact may be the main source of disagreement among observers, as it hinders the distinction between IEP with APFC and necrotizing pancreatitis with peripancreatic necrosis alone. This distinction is important, as the two conditions differ in their prognosis: a good outcome can be expected with IEP, whereas the prognosis of peripancreatic necrosis alone is more severe, although better than that of parenchymal necrosis [11, 12, 16]. Magnetic resonance imaging (MRI) has been found more sensitive than CT for detecting necrotic material [17]. In addition, interobserver agreement was better with MRI than with CT (κ = 0.469 versus κ = 0.257) for characterizing peripancreatic collections, particularly after the first month [18]. Previous work has established that the distinction between APFC and ANC is not reliable in the first days, leading to the suggestion that collections identified during this period should be designated Bindeterminate peripancreatic collections^ [14]. The type of collection seemed to be associated with clinical AP severity in our study. Thus, unsurprisingly, APFCs were more common in mild AP. ANCs were more common than APFCs in the groups with moderately severe and severe AP. In an earlier study [14], the association between AP type (IEP or parenchymal necrosis alone) and clinical outcome was stronger when the CTs were read by an expert instead of local radiologists. Interobserver agreement in our study was only moderate for detecting absence of parenchymal contrast enhancement. This sign, which indicates parenchymal necrosis, is often the key to diagnosing acute necrotizing pancreatitis and may help to characterize a peripancreatic collection, since any collection in a patient with parenchymal necrosis should be classified as an ANC [7]. However, unenhanced pancreatic areas larger than 3 cm or involving more than 30% of the gland are well known to be diagnosed with greater accuracy than smaller lesions [13], and in our study, parenchymal necrosis usually involved less than 30% of the gland. In the earlier study involving local radiologists [14], mean interobserver agreement was good for detecting parenchymal necrosis but the κ values varied Fig. 3 Acute peripancreatic fluid collections (APFCs): agreement between the two readers. Contrast-enhanced CT images at the portal phase in a 24-year-old man with AP. Axial (a, b) and coronal images (c) on day 1 show homogeneous peripancreatic collections, with fluid density, adjacent to the pancreas (arrows). The two readers agreed on the absence of parenchymal necrosis and on the classification of the collections Eur Radiol Fig. 4 Acute peripancreatic fluid collections (APFCs): disagreement between the two readers. Contrast-enhanced CT images at the portal phase on day 1 of AP in a 51-year-old woman. Axial images show homogeneous peripancreatic collections (arrows) extending toward the left anterior pararenal space. Both readers agreed on the absence of parenchymal necrosis. Reader 1 classified the collections as ANCs and reader 2 as APFCs. The consensus meeting with the expert radiologist led the collections to be classified as APFCs. Clinically, the AP was classified as moderately severe but the patient was discharged on day 3 widely across individual readers, from 0.319 to 0.731. In the future, the development of advanced techniques such as spectral imaging and perfusion imaging may improve agreement. For instance, perfusion CT obtained within 72 h of severe AP onset performed well in predicting parenchymal necrosis, with 87.5% sensitivity and 100% specificity compared to conventional contrast-enhanced CT obtained after 3 weeks [19]. Similar findings were reported in 2007 for perfusion CT [20] and in 2014 for subtraction CT mapping [21]. Dual-energy CT provides better contrast resolution that may help to detect parenchymal necrosis [22]. Agreement between readers was moderate for detecting parenchymal calcifications and pancreatic duct dilatation. These results may have been influenced by the low prevalence of both signs. Our study has several limitations. First, the design was retrospective. Second, no reference standard was used to determine the best terms for collections. Accurately determining the content of the collections (i.e., fluid and/or Fig. 5 Acute necrotic collections (ANCs): disagreement between the two readers. Contrast-enhanced CT images at the pancreatic phase (a, d, e) and portal phase (d, c) of AP in a 74-year-old man. On day 4, axial (a, b) and coronal images (c) show slightly heterogeneous, peripancreatic collections (arrows). Both readers agreed on the absence of parenchymal necrosis. Reader 1 classified the collections as ANCs and reader 2 as APFCs. The follow-up CT (d, e) performed 5 weeks after AP onset shows walled-off necrosis (arrows) indicating necrotizing AP Eur Radiol Informed consent Written informed consent was waived by the Institutional Review Board. Ethical approval Institutional Review Board approval was obtained. Methodology • retrospective • observational • performed at one institution Publisher’s Note Springer Nature remains neutral with regard to jurisdictionFig. 6 Parenchymal necrosis: disagreement between the two readers. Contrast-enhanced CT image at the portal phase on day 1 of AP in a 68-year-old woman. According to reader 1, there is parenchymal necrosis in the pancreas body (arrows). For reader 2, there is not parenchymal necrosis. The consensus meeting led to the presence of necrosis. Clinically, the AP was classified as mild, and the endoscopic ultrasonography shows parenchymal necrosis al claims in published maps and institutional affiliations. References 1. 2. solid) would have required routine aspiration, which would have been unethical, or MRI, which was not available at the time. Follow-up investigations are not a good reference standard in this case. For instance, resolution of a collection after 4 weeks does not prove that the collection was an APFC. Third, flares of AP in patients with chronic pancreatitis were not excluded. This point may have induced classification bias, as collections may have developed during earlier flares. However, as only 9% of our patients had chronic pancreatitis, any bias would not have substantially affected our findings. Fourth, 63 patients (35%) were examined by CT in the first 3 days, suggesting that selection bias may have occurred if early CT was performed because of more severe clinical picture. The recommended time for assessing AP by CT is 72 h after symptoms onset [6]. Earlier CT has low yield and no implications for clinical management [23]. In conclusion, our study shows a moderate interobserver agreement when using the RAC to categorize pancreatic and peripancreatic collections by CT, suggesting a need for either developing a new semiology to characterize peripancreatic collections by CT or for using another imaging modality such as MRI to better analyze collection contents. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. Funding The authors state that this work has not received any funding. 13. Compliance with ethical standards 14. Guarantor The scientific guarantor of this publication is Marc Zins. Conflict of interest The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. 15. 16. Statistics and biometry One of the authors has significant statistical expertise. Lankisch PG, Apte M, Banks PA (2015) Acute pancreatitis. Lancet 386(9988):85–96 Bradley EL 3rd (1992) A clinically based classification system for acute pancreatitis. Summary of the International Symposium on Acute Pancreatitis, Atlanta, Ga, September 11 through 13, 1992. Arch Surg 128:586–590 Besselink MG, van Santvoort HC, Bollen TL et al (2006) Describing computed tomography findings in acute necrotizing pancreatitis with the Atlanta classification: an interobserver agreement study. Pancreas 33:331–335 da Costa DW, Boerma D, van Santvoort HC et al (2014) Staged multidisciplinary step-up management for necrotizing pancreatitis. Br J Surg 101:e65–e79 Shyu JY, Sainani NI, Sahni VA et al (2014) Necrotizing pancreatitis: diagnosis, imaging, and intervention. Radiographics 34:1218–1239 Thoeni RF (2015) Imaging of acute pancreatitis. Radiol Clin North Am 53:1189–1208 Banks PA, Bollen TL, Dervenis C et al (2013) Classification of acute pancreatitis—2012: revision of the Atlanta classification and definitions by international consensus. Gut 62:102–111 Thoeni RF (2012) The revised Atlanta classification of acute pancreatitis: its importance for the radiologist and its effect on treatment. Radiology 262:751–764 Murphy KP, O’Connor OJ, Maher MM (2014) Updated imaging nomenclature for acute pancreatitis. AJR Am J Roentgenol 203: W464–W469 Forsmark CE, Vege SS, Wilcox CM (2016) Acute pancreatitis. N Engl J Med 375:1972–1981 Bakker OJ, van Santvoort H, Besselink MG et al (2013) Extrapancreatic necrosis without pancreatic parenchymal necrosis: a separate entity in necrotising pancreatitis? Gut 62:1475–1480 Wang M, Wei A, Guo Q et al (2016) Clinical outcomes of combined necrotizing pancreatitis versus extrapancreatic necrosis alone. Pancreatology 16:57–65 Bollen TL (2016) Acute pancreatitis: international classification and nomenclature. Clin Radiol 71:121–133 Sternby H, Verdonk RC, Aguilar G et al (2016) Significant interobserver variation in the diagnosis of extrapancreatic necrosis and type of pancreatic collections in acute pancreatitis - an international multicenter evaluation of the revised Atlanta classification. Pancreatology 16:791–797 Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33:159–174 Singh VK, Bollen TL, Wu BU et al (2011) An assessment of the severity of interstitial pancreatitis. Clin Gastroenterol Hepatol 9: 1098–1103 Eur Radiol 17. Morgan DE, Baron TH, Smith JK, Robbin ML, Kenney PJ (1997) Pancreatic fluid collections prior to intervention: evaluation with MR imaging compared with CT and US. Radiology 203:773–778 18. Kamal A, Singh VK, Akshintala VS et al (2015) CT and MRI assessment of symptomatic organized pancreatic fluid collections and pancreatic duct disruption: an interreader variability study using the revised Atlanta classification 2012. Abdom Imaging 40: 1608–1616 19. Yadav AK, Sharma R, Kandasamy D et al (2015) Perfusion CT: can it predict the development of pancreatic necrosis in early stage of severe acute pancreatitis? Abdom Imaging 40:488–499 20. Tsuji Y, Yamamoto H, Yazumi S et al (2007) Perfusion computerized tomography can predict pancreatic necrosis in 21. 22. 23. early stages of severe acute pancreatitis. Clin Gastroenterol Hepatol 5:1484–1492 Tsuji Y, Takahashi N, Fletcher JG et al (2014) Subtraction color map of contrast-enhanced and unenhanced CT for the prediction of pancreatic necrosis in early stage of acute pancreatitis. AJR Am J Roentgenol 202:W349–W356 Marin D, Boll DT, Mileto A, Nelson RC (2014) State of the art: dual-energy CT of the abdomen. Radiology 271:327–342 Spanier BW, Nio Y, van der Hulst RW, Tuynman HA, Dijkgraaf MG, Bruno MJ (2010) Practice and yield of early CT scan in acute pancreatitis: a Dutch Observational Multicenter Study. Pancreatology 10:222–228