Comparison of imaging modalities for measuring the diameter of intraductal papillary mucinous neoplasms of the pancreas

Abstract

BACKGROUND:

Intraductal papillary mucinous neoplasms (IPMNs) are pre-malignant pancreatic cysts detected by imaging. Cyst size is one of many features evaluated on computed tomography (CT), magnetic resonance imaging (MRI), or endoscopic ultrasonography (EUS) to help guide IPMN management. Our objective was to determine which imaging modality best predicts pathological cyst size.

METHODS:

We analyzed records for 57 IPMN cases surgically treated at Moffitt Cancer Center from 2008-2016 for whom pre-operative CT, MRI, and EUS IPMN cyst size and post-operative pathological cyst size values were available. Long axis cyst diameter measurements were compared to each other and corresponding pathological cyst measurements using within-subjects ANOVA, Bland-Altman analysis, and linear regression. Consensus measurements were also performed on CT and MRI images.

RESULTS:

Cyst size measured via CT and MRI overestimated pathological size by 0.33 cm and 0.27 cm, respectively, whereas EUS underestimated pathological size by 0.05 cm and had the narrowest 95% limit of agreement (LOA). Among pathologically-confirmed cysts < 3 cm, MRI overestimated pathological size by 0.30 cm (P=0.049) and had the widest LOA, followed by EUS and CT. Among cysts ≥3 cm, EUS underestimated pathological size by 0.35 cm (P=0.059) and MRI and CT overestimated pathological size by 0.23 cm and 0.51 cm, respectively.

CONCLUSIONS:

In this small retrospective study, EUS cyst size measurements correlated best with pathologic specimens compared to CT and MRI, especially for cysts < 3 cm. Larger prospective studies are needed to determine which imaging modalities are best to risk-stratify IPMNs and guide surgical versus. non-surgical management.

Introduction

With advances in the quality and widespread use of abdominal imaging in the aging population, pancreatic cystic lesions (PCL) are being incidentally detected in 3% of all computed tomography (CT) scans and 20% of all magnetic resonance imaging (MRI) scans each year ^{(1, 2)}. The radiologic detection of these so-called pancreatic incidentalomas is not trivial since they can be benign, premalignant, or malignant, with 40-70% representing pre-malignant lesions known as intraductal papillary mucinous neoplasms (IPMNs) ^{(1, 3, 4)}. To aid in decision-making regarding the option to surgically resect or monitor IPMNs, international consensus guidelines (ICG) ^(5-8) have been developed which rely on conventional radiologic features. Cyst size is an important feature that can be evaluated on CT, MRI, or endoscopic ultrasound (EUS) to aid in decision making. Current ICG recommend consideration of surgery when cyst size is at least 3 cm while close surveillance by CT or MRI is recommended for cysts smaller than 3 cm (assuming no ‘high risk stigmata’ such as main duct (MD) involvement/dilatation ≥10 mm, obstructive jaundice, or an enhanced solid component/mural nodule ≥5 mm are present). Additionally, cyst growth of >2 mm per year is also thought to be a predictor of malignancy ⁽⁹⁾. Recent studies from leaders in the field suggest ICG accurately stratify IPMNs for pancreatic cancer risk, with cysts that screen negative for ICG having the lowest risk and cysts that screen positive for ICG having the greatest risk for pancreatic cancer development⁽¹⁰⁾ and/or a higher risk of a subsequent pancreatic cancer post resection via a partial pancreatectomy⁽¹¹⁾.

Cyst size is critical to consider when recommending surgical resection or surveillance. However, consistent with prior studies that noted discrepancies when comparing radiological tumor size with pathological tumor size in breast and renal tissues ^{(12, 13)}, limited studies of pancreatic tumors also demonstrate the inability of radiographic measurements to adequately correlate with final pathological tumor size ^(14-16). In particular, PCL size has been shown to be discordant between imaging modalities (CT, MRI, and EUS), and measurement of cyst size via pre-operative imaging often does not correlate well with the size of the pathologic specimen after resection^{(15, 16)}. For example, in a Korean study of 34 PCL cases (including 15 IPMNs), Lee et al. reported EUS had the lowest level of agreement, followed by MRI, with CT showing the highest level of agreement among all three modalities⁽¹⁶⁾. In contrast, in a study of pancreatic cancers (not cysts), Arvold et al. reported CT imaging significantly under-represented pancreatic cancer size in resectable cases ⁽¹⁴⁾. Given that pancreatic surgical resections can be associated with mortality rates of approximately 5% and morbidity rates between 30% and 60%^{(9, 17, 18)}, the diagnostic accuracy of pre-operative cyst size should be strongly considered in deciding how and when to treat IPMNs.

The main objective of the current investigation was to conduct the first study we are aware of in the United States to specifically evaluate the diagnostic accuracy of CT, MRI, and EUS in predicting pathological tumor size specifically in a retrospective cohort of IPMN cases. Furthermore, we also examined whether discrepancies exist within clinically relevant subgroups based on size (<3 cm vs. ≥3 cm), cyst location (head vs body/tail), and main duct involvement (main/mixed vs branch duct).

Methods

We retrospectively reviewed clinicopathologic findings and radiologic images for patients diagnosed and treated at Moffitt Cancer Center (Tampa, Florida) between 2008 and 2016 who had consented to an Institutional Review Board-approved protocol. Fifty-seven surgically-resected, pathologically-confirmed IPMN cases (30 male, 27 female) were identified for whom preoperative CT, MRI, and EUS cyst sizes as well as post-operative pathological cyst sizes were available. CT and MRI cyst size measurements were performed by board certified radiologists using the linear measurement tool on GE Healthcare Centricity Universal Viewer Picture Archiving Communication (PACS) version 6.0 (GE Healthcare, Waukesha WI, USA). EUS long axis cyst size measurements were performed using the measurement tool on the Aloka Alpha 5 prosound processor. For patients with multifocal cysts, the largest cyst qualified as the primary index cyst or was excluded from the study if size could not be verified or tracked across imaging modalities. No cases received prior neoadjuvant therapy. Images for a representative IPMN case from our cohort are shown in Figure 1. The primary endpoint for radiologic and pathological size was the maximum diameter in centimeters (cm) of the primary cyst in any direction. Secondary endpoints included median time from imaging modality to surgical resection, location in the pancreas, branch or main duct communication, and IPMN grade. Pathological size was determined at the time of frozen section analysis of the surgical margin and was used as the reference standard.

Figure 1:

A) Axial arterial phase post-contrast CT demonstrates a hypoenhancing uncinate process lesion. B) Axial T2 weighted MRI shows corresponding heterogeneous high T2 signal suggesting cystic fluid contents. C) Coronal MRCP chest image shows the multiloculated nature of the mass. D) Endoscopic ultrasound image demonstrates multiplecystic loculations with internal thick septations. Red arrows indicate mass boundaries.

Consensus measurements were performed on all available CT and MRI exams. The maximum cyst diameter was measured using GE Healthcare Centricity Universal Viewer Picture Archiving Communication (PACS) version 6.0 (GE Healthcare, Waukesha WI, USA) for each modality. Axial and coronal post contrast venous phase images viewed at full screen were used for CT measurements and the largest single diameter was recorded as the CT cyst size. Axial and coronal T2 weighted non-fat saturated images also viewed at full screen were used for MRI measurements and the largest single diameter was recorded as the MRI cyst size. The consensus observer was blinded to cyst sizes for these measurements.

Statistical analysis was performed using within-subjects ANOVA followed by Tukey’s Honest Significant Difference (HSD) test between the difference of long axis cyst diameter on CT, MRI, and EUS with corresponding pathological cyst size. We also calculated intraclass correlation coefficient (ICC), 95% Limit of agreement (LOA), and 95% Confidence intervals and constructed Bland-Altman plots using MedCalc statistical software (MedCalc Software, Ostend, Belgium). Pearson’s correlation was calculated to compare CT, MRI, and EUS cyst size with pathological size. The differences were calculated for the cohort overall. ICC was also used to calculate consensus agreement for CT and MRI exams. Stratification was also performed based on pathological cyst size <3.0 cm and ≥3.0cm. Box-and-Whisker plots were constructed from the difference between imaging modality and pathological size for pancreatic cyst location and type of duct involvement for all 3 imaging modalities: CT, MRI, and EUS.

Results

Study population characteristics

The clinical and pathological characteristics of the 57 patients included in this study are summarized in Table 1. The mean age was 70.9, and the majority of cases (52.6%) were male. Most cases (n=38, 66.7%) had cysts in the pancreatic head and most tumors (n=41, 71.9%) only involved the side branch ducts. The majority of resected cases had invasive or high-grade dysplasia (Table 1).

Table 1:

Clinical and pathological characteristics of IPMN cases in the cohort (n=57)

Variable
Mean age, years (range; SD)	70.9 (45-83; 8.9)
Sex, n (%)
Male	30 (52.6)
Female	27 (47.4)
Location in pancreas, n (%)
Head	38 (66.7)
Body	11 (19.3)
Tail	8 (14.0)
Branch Communication, n (%)
Main	3 (5.3)
Side	41 (71.9)
Mixed	13 (22.8)
IPMN Grade, n (%)
Low	4 (7.0)
Moderate	13 (22.8)
High	19 (33.3)
Invasive	21 (36.8)

Comparison of radiological and pathological tumor size

Long axis cyst size measured via CT scan (mean=3.15 cm, SD=1.84) and MRI (mean=3.09 cm, SD=1.76) significantly overestimated pathological size (mean=2.82 cm, SD=1.25) by 0.33 cm (P=0.017) and 0.27 cm (P=0.038), respectively (Table 2). This trend was not observed with EUS. For cysts ≥ 3 cm, CT scans overestimated pathological size 0.51 cm (P=0.046) (Table 2). EUS significantly underestimated pathological size by 0.35 cm (P=0.059) for cysts ≥3 cm and MRI overestimates pathological cyst size by 0.30cm (P=0.049) for cysts <3 cm (Table 2). The median time between imaging modality and surgical resection was 35 days for CT, 87 days for MRI, and 65 days for EUS (Table 2). The ICC ((95% CI) was 0.95 (0.92-0.97)] for CT exams, and 0.92 (0.86-0.95)] for MRI exams.

Table 2:

Mean radiological and pathological tumor size stratified by modality and clinically relevant size subgroups

Imaging modality	Pathological size range (cm)	n	Median time between modality (days)	Mean imaging diameter (SD) (cm)	Mean pathological diameter (SD) (cm)	Difference	P
CT Scan	Overall	57	35	3.15 (1.84)	2.82 (1.25)	0.33	0.017
	<3	32	32	2.25 (0.89)	2.07 (0.56)	0.18	0.199
	≥3	25	35	4.29 (2.04)	3.78 (1.21)	0.51	0.046
MRI	Overall	57	87	3.09 (1.76)	2.82 (1.25)	0.27	0.038
	<3	32	72	2.37 (0.98)	2.07 (0.56)	0.30	0.049
	≥3	25	78	4.00 (2.04)	3.78 (1.21)	0.23	0.310
EUS	Overall	57	65	2.77 (1.14)	2.82 (1.25)	−0.05	0.690
	<3	32	69	2.26 (0.95)	2.07 (0.56)	0.19	0.202
	≥3	25	56	3.43 (1.01)	3.78 (1.21)	−0.35	0.059

Note: The statistical test performed was a one-way within subjects analysis of variance (ANOVA) on the cyst diameter size between CT,MRI, and EUS pathology.

Differences in size estimates between each imaging modality and the resected specimen

Overall, MRI showed the widest range of limit of agreement (LOA) between all 3 imaging modalities from −1.59 to +2.12. CT Scans showed the second widest range of LOA between −1.63 to +2.29 followed by EUS which had the narrowest range of LOA between −1.78 to +1.69 (Table 3). The intraclass correlation coefficient (ICC) was 0.787, 0.802, and 0.735 for CT, MRI, and EUS, respectively, demonstrating all 3 imaging studies had excellent reliability with pathological size overall (Table 3). Linear regression performed on Bland-Altman plot suggests as pathological size increases, CT (r=0.33, P=0.011) and MRI (r=0.28, P=0.034) tend to overestimate the actual size of the IPMN while EUS tends to underestimate the actual size (r=−0.47, P=0.0003) (Figure 2).

Table 3:

Differences in Size Estimates Between Each Imaging Modality and the Resected Pathological Specimen

		CT vs Pathology	MRI vs Pathology	EUS vs Pathology
Overall	Mean Difference (SD) (cm)	0.33 (1.00)	0.27 (0.95)	−0.05 (0.88)
	95% Limit of agreement	−1.63 to +2.29	−1.59 to +2.12	−1.78 to +1.69
	ICC	0.787	0.802	0.735
	95% Confidence Interval	0.653 to 0.871	0.681 to 0.879	0.588 to 0.835
	Subgroup Bland-Altman Analysis between Imaging Modality and Pathology for IPMN <3 cm
		CT vs Pathology	MRI vs Pathology	EUS vs Pathology
<3 cm	Mean Difference (SD) (cm)	0.18 (0.78)	0.30 (0.82)	0.19 (0.82)
	95% Limit of agreement	−1.35 to +1.71	−1.32 to +1.91	−1.42 to +1.80
	ICC	0.465	0.465	0.457
	95% Confidence Interval	0.151 to 0.695	0.156 to 0.694	0.142 to 0.690
	Subgroup Bland-Altman Analysis between Imaging Modality and Pathology for IPMN ≥3 cm
		CT vs Pathology	MRI vs Pathology	EUS vs Pathology
≥3 cm	Mean Difference (SD) (cm)	0.51 (1.22)	0.23 (1.10)	−0.35 (0.84)
	95% Limit of agreement	−1.88 to +2.90	−1.96 to +2.39	−2.08 to +1.38
	ICC	0.731	0.799	0.686
	95% Confidence Interval	0.469 to 0.874	0.601 to 0.906	0.403 to 0.848

Abbreviations: ICC=intraclass correlation coefficient; SD=standard deviation; CT=computed tomography; MRI=magnetic resonance imaging; EUS=endoscopic ultrasound

Figure 2:

Bland-Altman plot between cyst size on pathology and A) CT, B) MRI, and C) EUS, with mean difference and 95% limits of agreement.

Sub-group analyses by cyst size, location, and main duct involvement

For pathological cyst sizes <3 cm, the ICC for CT, MRI, and EUS were below 0.5, which is indicative of poor reliability (CT=0.465, MRI=0.465, EUS=0.457) (Table 3). In pathological cysts <3 cm in size, MRI consistently overestimated the pathological size by 0.30 cm (P=0.049) (Table 2). Bland-Altman analysis and linear regression reveals no discernable bias between CT size and pathological size (r=−0.128, P=0.985) (Figure 3A). The difference between cyst size on MRI and EUS as pathological size increases is not significant (r=0.009, P=0.961: r=0.080, P=0.663) (Figure 3A). The mean difference between imaging diameter and pathological diameter for cysts <3 cm was 0.30 cm in MRI, 0.19 cm in EUS and 0.18 cm in CT scans (Table 2). MRI showed the widest range of LOA followed by EUS, while CT had the narrowest range.

Figure 3:

Bland-Altman plot between pathologic cyst size and size on CT, MRI, and EUS, with mean difference and 95% limits of agreement A) for cysts <3.0 cm and B) for cysts ≥3.0 cm.

In pathological cysts ≥3 cm in size, CT consistently overestimated the pathological size by 0.51 cm (P=0.046) (Table 3). Additionally, EUS consistently underestimated the pathological size by 0.35 cm (P=0.059) (Table 3). Linear regression performed on Bland-Altman reveal the difference between CT and pathological size and MRI and pathological size appear to increase as pathological size increases (r=0.475, P=0.016; r=0.622, P=0.0009) (Figure 3B). Furthermore, as pathological size increases, the difference between EUS size and pathological size is negatively correlated (r=−0.562, P=0.0035) suggesting that underestimation of pathological significantly increases. (Figure 3B). The mean difference between pathological diameter and imaging diameter for cysts ≥3 cm was 0.51cm in CT, 0.23 cm in MRI, and −0.35cm in EUS (Table 3). CT showed the widest range of LOA followed by EUS, while MRI had the narrowest range.

There was no significant difference between cyst size on CT, MRI, and EUS versus pathological size when stratified by tumor location in the head, body, or tail (Figure 4A). However, CT scans significantly overestimated main duct IPMNs when compared to side branch and mixed IPMNs (P=0.002, P=0.003) (Figure 4B). This trend was not observed on MRI or EUS.

Figure 4:

Box-and-whisker plot of cyst size difference between CT and pathology, MRI and pathology, and EUS and pathology based on A) cyst location in the pancreas, and B) pancreatic main duct involvement. Statistical tests performed included one-way ANOVA followed by Tukey’s Honest Significant Difference (HSD) test for head/body/tail and for main duct/side branch/mixed type.

Discussion

Consensus is lacking regarding the imaging modality that best predicts pathologic size of IPMNs. Findings from the current study reveal that IPMN size measured by CT and MRI is significantly larger than final pathological cyst size, while EUS cyst size correlates well with pathologic size, especially for cysts <3 cm. These findings are in line with those of other studies which reported an overestimation of tumor size by CT scan among non-pancreatic ^(11,15) and pancreatic lesions ⁽¹⁴⁾ . However, our findings differ from a smaller study by Lee et al. which showed that EUS significantly underestimated pathological size, particularly when the tumor was located in the pancreatic head ⁽¹⁶⁾. In the current study, cyst size measurement did not seem to vary by tumor location. One possible explanation for the large discrepancy between CT and MRI sizes compared to EUS size is due the visualization of lesions in different planes. While CT and MRI typically use axial or coronal cuts, EUS images lesions at oblique angles which can cause some variation⁽¹⁶⁾. EUS allows for direct visualization of the lesion, higher resolution and resulting sensitivity for detecting pancreatic neoplasms, as well as the ability to perform tissue biopsy or fluid aspiration to increase diagnostic accuracy^{(18, 19)}. One important note is that although EUS cyst size correlates well with pathologic size, there was a high percentage of cysts with high grade dysplasia or carcinoma on resection among cysts that were smaller on EUS. For example, 67.9% of cysts that were smaller on EUS compared to pathological size had high-grade dysplasia or invasive carcinoma (compared to 68.4% and 61.9% of cysts that were smaller on MRI and CT compared to pathologic size, respectively) (Table 4).

Table 4:

IPMN grade stratified by undersized cysts or oversized cysts on imaging

	Undersized on Imaging
	Low Grade Dysplasia	Moderate Dysplasia	High Grade Dysplasia	Invasive Carcinoma
CT Scan (n=21)	1	7	6	7
MRI (n=19)	2	3	7	6
EUS (n=28)	3	6	11	8
	Oversized on Imaging
	Low Grade Dysplasia	Moderate Dysplasia	High Grade Dysplasia	Invasive Carcinoma
CT Scan (n=35)	3	6	12	14
MRI (n=30)	2	7	10	12
EUS (n=27)	1	6	8	12

One major weakness of the current study is the small sample size and retrospective nature. Additionally, similar to other retrospectively conducted investigations in this topic area, there was no standardized protocol for examining radiographic or pathological tumor size. The cohort had a range of CT scans including thorax, abdomen, pelvis, and pancreatic protocol with and without contrast. Another concern is the time between imaging modality and surgical resection. The median time between CT and surgery was 35 days compared to MRI and EUS which had median times of 87 days and 65 days, respectively. Cyst aspiration during EUS may have also impacted cyst size measurements for cases when EUS was performed after CT and MRI but before surgical resection. The mean growth rate of IPMNs is reported to be between 0.08 cm and 1.86 cm per year, with malignant IPMNs growing at a faster rate than non-malignant IPMNs^{(9, 20)}. The EUS procedure also introduces variability in measurements as the endoscopic probe position, angulation, and probe pressure within the duodenum may vary slightly between patients. Additionally, the cyst measurements on CT and MRI were performed as standard of care and the longest of 3 axis measurements was used. The specific MRI sequence that produced the largest cyst diameter was utilized for measurement, however this also could have varied between patients. Consensus measurements suggest that CT had better interobserver agreement than MRI which may be related to the increased number of potential images and planes seen with each MRI exam compared to corresponding CT exams. Beyond selecting the maximum cyst diameter, there is no universally accepted standardized imaging sequence or plane to best measure pancreatic cysts.

Another limitation to consider pertains to measurement of final tumor/pathological staging and measurements. Tran et al. reported fresh specimens were 4.6% larger than formalin-fixed specimens and that formalin-fixed specimens were 7.1% larger than microscopic measurements^[18]. However, a shrinkage factor for fresh tumors can be accounted for by using correction factors based on whether the tumor is fixed or on a slide (1.05 and 1.08)^{(21, 22)}. Tumor shrinkage post-resection is consistent within pathological tissue processing laboratories within the same institution and can therefore still be used to measure pathological tumor size with high internal reliability⁽²³⁾. Another possible limitation is the fact that pathological measurements are performed in 2 dimensions after the cyst is opened in contrast to clinical imaging studies that are performed with an intact cyst wall in 3 dimensions. It is likely that measurement of an open cyst may underestimate the true size of the intact cyst.

Despite these limitations, this is the first study to compare cyst size within a relatively homogenous group of patients whose IPMNs were measured preoperatively by CT, MRI, and EUS and post-operatively at the time of pathologic evaluation. This investigation provides support that EUS may be a superior method for measuring the size of IPMNs. However, other factors that may influence a clinician’s decision for a specific type of imaging such as the ability to characterize other morphological features or accurately stage tumors must be considered in conjunction with other radiologic and clinical features for proper risk assessment and counseling regarding management. This work confirms the variability of cyst size based on imaging modality and supports the recommendation that active surveillance of pancreatic neoplastic cysts should be performed with a single imaging modality to guide clinical management based on changes in cyst size.