Endoscopic Ultrasound-Based Differentiation of Pancreatic Neuroendocrine Tumors and Pancreatic Duct Adenocarcinoma

Summary

This study outlines a method using digital image analysis to differentiate between pancreatic neuroendocrine neoplasms (PNEN) and pancreatic ductal adenocarcinoma (PDAC).

Abstract

Endoscopic ultrasound (EUS) is one of the most critical imaging modalities for evaluating pancreatic masses, providing high-resolution visualization of the pancreas. This study aimed to assess the feasibility of digital image analysis (DIA) in differentiating pancreatic neuroendocrine neoplasms (PNEN) from pancreatic ductal adenocarcinoma (PDAC) in solid pancreatic tumors observed via EUS. A retrospective analysis was conducted on all solid pancreatic tumors examined by EUS procedures from 2018 to 2023, including cases with confirmed pathological diagnoses of PDAC and PNEN. Adobe Photoshop was used to perform DIA. The gray ratio, defined as the gray-scale ratio of the lesion to a selected background, was calculated to ensure consistency. Circularity was determined by the ratio of area to perimeter. Additionally, gray-scale standard deviations were extracted for further analysis. Three quantitative parameters from EUS images demonstrated potential for distinguishing PNEN from PDAC: circularity, gray ratio, and gray-scale standard deviation. Specifically, a circularity greater than 0.68, a gray ratio exceeding 0.40, and a gray-scale standard deviation below 12.44 were indicative of potential PNEN.

Introduction

Pancreatic cancer consists of both solid and cystic lesions and ranks among the top ten leading causes of cancer-related death worldwide¹. The most prevalent histological type of solid pancreatic tumor is pancreatic ductal adenocarcinoma (PDAC), accounting for 95% of malignant pancreatic tumors. At the time of diagnosis, approximately 80% of PDAC cases present with locally advanced disease or metastasis, with a median survival time of less than one year². Surgical and postoperative adjuvant therapies are feasible for only the remaining 20% of patients and may extend the 5-year survival rate to approximately 10%-22%.

Pancreatic neuroendocrine neoplasms (PNEN) represent the second most common solid pancreatic tumors and have a better prognosis than PDAC³. PNENs are classified as functional or nonfunctional based on the secretion of pancreatic endocrine hormones and associated clinical symptoms. Treatment strategies for PNEN differ significantly from those for PDAC. For early-stage PNEN, surgery is the primary recommendation and remains the only "curative" treatment available. Additionally, chemotherapy for PNEN is more specific and selective compared to PDAC, utilizing agents such as somatostatin analogs and receptor tyrosine kinase inhibitors. According to the WHO classification, PNENs are divided into three histological grades based on mitotic count and Ki-67 index, both of which serve as reliable prognostic indicators⁴.

Given the differences in epidemiology, prognosis, and treatment between PDAC and PNEN, precise diagnosis is crucial for effective treatment. In the management algorithm for pancreatic tumors, endoscopic ultrasound (EUS) and EUS-guided fine needle aspiration (EUS-FNA) have become core modalities for diagnosis, staging, and treatment. Lesions with irregular edges and pancreatic duct dilation on EUS are often indicative of PDAC rather than PNEN^5,6,7. However, the interpretation of ultrasound images-a key component of EUS-can be subjective and variable, depending on the operator's experience.

Recently, digital imaging analysis (DIA) has been increasingly utilized in medical diagnostics, including applications in computed tomography (CT) and ultrasonography⁸. The characteristics of masses are captured by the probes and converted into pixel information, which forms the gray-scale imaging available for further computer analysis. Kumon et al. were pioneers in applying DIA to differentiate between normal and diseased pancreas and lymph nodes⁹. Subsequently, other researchers explored the use of DIA for the differential diagnosis of pancreatic cancer from normal tissue, employing a neural-network-based predictive model and support vector machine theory⁹.

Building on these advancements, the current study aimed to assess whether DIA could effectively distinguish pancreatic neuroendocrine neoplasms (PNEN) from pancreatic ductal adenocarcinoma (PDAC) in solid pancreatic tumors using endoscopic ultrasound (EUS).

Protocol

The institutional review board at Ruijin Hospital approved the study [Clinical Ethics Review No. 75 (2012)], and the protocol adhered to the institutional guidelines for human care. All pancreatic lesions examined via endoscopic ultrasound (EUS) at Ruijin Hospital from 2018 to 2023 were retrospectively analyzed. Data were retrieved from the electronic medical database.

The inclusion criteria were: (1) EUS imaging showed a distinct lesion; (2) Histological diagnosis of pancreatic ductal adenocarcinoma (PDAC) or pancreatic neuroendocrine neoplasm (PNEN) was obtained. The exclusion criteria were as follows: (1) Incomplete or partially missing data; (2) Tumors surrounding the portal vein, splenic vein, mesenteric arteriovenous structures, ampulla, and cavitary tissues that could affect image analysis; (3) Ultrasound imaging affected by stent implantation in the bile or pancreatic ducts; (4) Digital images of endoscopic ultrasonography not available from the hospital; (5) Unclear postoperative pathological examination results or findings not corresponding to PNEN or PDAC; (6) EUS images displaying non-solid pancreatic spaces, such as cystic or combined cystic lesions. All EUS procedures were performed by two experienced endosonographers using a linear scanning echo-endoscope operating at frequencies of 5 MHz or 7.5 MHz. The EUS features of the target areas were documented, and at least five corresponding images were captured during a comprehensive examination of the entire pancreas and adjacent organs. Informed consent was obtained from all patients, who received conscious sedation with intramuscular diazepam. The reagents and equipment used in the study are listed in the Table of Materials.

1. Preoperative preparation

1. Ensure the patient fasts for at least 12 h prior to the procedure.
2. Administer general intravenous anesthesia before starting the procedure (following institutionally approved protocols).

2. Gastroscopic exploration

1. Insert the endoscope through the patient's mouth and inflate the gastric cavity with gas.
2. Gradually advance the endoscope along the esophagus until reaching the entrance of the duodenal papilla.
3. Insert the ultrasound probe into the endoscope and position it near the target organ for imaging.
   NOTE: Ensure that two experienced gastroenterologists with over 5 years of experience perform the endoscopic ultrasound.

3. Duodenal scan of the pancreas

1. Insert the endoscopic ultrasound probe at the level of the duodenal papilla and adjust the bending knob to straighten the probe.
2. Remove any air and mucus from the duodenum. Fill the water bag with 5-15 mL of water to ensure close contact between the water bag wall and the duodenum.
3. Adjust the endoscope's left-right and up-down control buttons. Carefully maneuver the ultrasound endoscope externally to maintain optimal imaging.
   NOTE: If gas interference is observed in the intestinal cavity after the ultrasound image is displayed, inject deaerated water through the biopsy forceps channel and adjust the probe position to eliminate the gas interference.

4. Gastrointestinal scan of the pancreas

1. Aspirate the deaerated water from the water bag and retract the endoscope to the antrum of the stomach after completing the duodenal scan.
2. Inject deaerated water into the water bag and retract the endoscopic ultrasound while displaying the ultrasound image.
3. Visualize the body and tail of the pancreas upon reaching the body and fundus of the stomach.
4. Inject 200-300 mL of deaerated water into the stomach to enhance the visibility of the pancreas body and tail.

5. Postoperative recovery

1. Collect satisfactory image data during the procedure and save the images to the server for future reference.
2. Remove the ultrasound probe and endoscope from the patient's body.

6. Image analysis

1. Store the digital images as 8-bit images with a resolution of 768 × 576 pixels in the Picture Archiving and Communication Systems (PACS).
2. Conduct imaging analysis using an image editing software.
3. Manually outline the distinct tumor area for Region of Interest (ROI) selection.
4. Use the Magic Wand tool to automatically select the ROI.
5. Utilize the Histogram and Analysis tools to analyze and record imaging parameters, including area, perimeter, width, height, major and minor diameters, roundness (reflecting tumor size and shape), average gray-scale value and gray-scale ratio (indicating intensity of echoes), and gray-scale standard deviation (assessing echo uniformity).
6. Delineate three ring-shaped ROIs, each with identical pixel counts and normal EUS appearances, to calculate the mean background density under standard operating conditions.
7. Calculate the gray-scale ratio by dividing the tumor's gray-scale by that of the background. Define circularity as the ratio of area to perimeter using the following formula¹⁰:
   Circularity = 4π × Area / (Perimeter²)
   NOTE: The gray-scale of the EUS image may vary depending on the contrast and gain settings during the examination. A flow chart illustrating these steps is shown in Figure 1.

7. Data analysis

1. Present quantitative data (Area, Perimeter, Circularity, Gray ratio, Major diameter, Gray-scale standard deviation, and age) as mean ± standard deviation (m ± SD). Compare these using the independent sample t-test or the Mann-Whitney U test.
2. Express qualitative data (Stage and Gender) as frequencies. Compare them using the Fisher exact test or the Pearson chi-square test. Consider a p-value of less than 0.05 as statistically significant.
3. Generate receiver operating characteristic (ROC) curves for major diameter, circularity, gray ratio, and gray-scale standard deviation. Calculate the cut-off values for each parameter.
4. Assess diagnostic value by calculating sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. If any parameter is negative, classify the combined result as negative.

Results

Patients' characteristics
A total of 119 patients underwent EUS puncture and histological diagnosis, comprising 63 patients with PDAC and 54 with PNEN. The average age of the participants was 55.4 years ± 14.10 years, with 47.86% identified as male. Based on EUS features, 59 cases were located in the pancreatic head, 58 in the body or tail, and three cases involved the entire pancreas. According to the WHO staging system, more than half of the PNEN cases were classified as Stage I (62.96%, 34/54). Fo...

Discussion

EUS is one of the most important imaging modalities for evaluating pancreatic masses, as it provides high-resolution visualization of the pancreas and enables the acquisition of adequate cytological samples through fine needle aspiration (FNA). Over the past decades, most studies have focused on improving the performance of FNA, while investigations into the echoic features of EUS imaging have remained limited. To date, studies evaluating the diagnostic value of EUS alone for differentiating PNEN in solid pancreatic tumo...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
22G Flexible Endoscopic Ultrasound Fine Needle Aspiration (EUS FNA) Needle	Boston Scientific International Medical Trading (Shanghai) Co., Ltd.	M00550010 (Single), M00550011 (Box of 5)
IBM SPSS Statistics version 22.0	IBM Corp, Armonk, NY	CRNP3ML, CIP2XML, CRNM6ML (various editions)
Linear scanning echo-endoscope, EG-530UT	Fujifilm (China) Investment Co., Ltd	EG-530UT; Balloon Pack Code: B20UT
Physiological saline (0.9% NaCl)	Chenxin Pharmaceutical Co., Ltd	Not listed (standard pharmaceutical supply)
Simethicone	Berlin-Chemie AG, Germany	Espumisan 100 mg/mL (no public catalog number)
Sodium hyaluronate	Shanghai Haohai Biotechnology Co., Ltd	Matrifill (no public catalog number)