Differential Diagnosis of Pancreatic Serous Cystadenomas and Mucinous Cystadenomas: Utility of Texture Analysis in Combination with Morphological Characteristics

Background Texture analysis of medical images has been reported to be a reliable technique for differential diagnosis of neoplasms. This study was to investigate the performance of textural features and the combined performance of textural features and morphological characteristics in the differential diagnosis of pancreatic serous and mucinous cystadenomas. Methods We retrospectively reviewed 59 patients with pancreatic serous cystadenoma and 32 patients with pancreatic mucinous cystadenoma at our hospital. Textural features were extracted using the LifeX software, and the least absolute shrinkage and selection operator (LASSO) method was applied to select the textural features. The differential diagnostic abilities of morphological features, textural features, and their combination were evaluated using receiver operating characteristic (ROC) analysis, with the area under the receiver operating characteristic curve (AUC) as the main indicator. Results The combination of morphological characteristics and textural features showed a higher AUC (0.893, 95% CI 0.816-0.970) than use of morphological characteristics (0.783, 95% CI 0.665-0.900) or textural features (0.777, 95% CI 0.673-0.880) alone. Conclusions In conclusion, our preliminary results highlighted the potential of CT texture analysis to discriminate pancreatic serous cystadenomas from mucinous cystadenomas. Furthermore, the combination of morphological and textural features can significantly improve the diagnostic performance, which may provide a reliable method to select patients with indication of surgical intervention in consideration of the different treatment principles of the two diseases.


Abstract
Background Texture analysis of medical images has been reported to be a reliable technique for differential diagnosis of neoplasms. This study was to investigate the performance of textural features and the combined performance of textural features and morphological characteristics in the differential diagnosis of pancreatic serous and mucinous cystadenomas. Methods We retrospectively reviewed 59 patients with pancreatic serous cystadenoma and 32 patients with pancreatic mucinous cystadenoma at our hospital. Textural features were extracted using the LifeX software, and the least absolute shrinkage and selection operator (LASSO) method was applied to select the textural features. The differential diagnostic abilities of morphological features, textural features, and their combination were evaluated using receiver operating characteristic (ROC) analysis, with the area under the receiver operating characteristic curve (AUC) as the main indicator. Results The combination of morphological characteristics and textural features showed a higher AUC (0.893, 95% CI 0.816-0.970) than use of morphological characteristics (0.783, 95% CI 0.665-0.900) or textural features (0.777, 95% CI 0.673-0.880) alone. Conclusions In conclusion, our preliminary results highlighted the potential of CT texture analysis to discriminate pancreatic serous cystadenomas from mucinous cystadenomas. Furthermore, the combination of morphological and textural features can significantly improve the diagnostic performance, which may provide a reliable method to select patients with indication of surgical intervention in consideration of the different treatment principles of the two diseases.

Background
Cystic neoplasms of the pancreas are historically considered rare subsets of pancreatic lesions. However, pancreatic neoplasms are being diagnosed more frequently given the widespread use of abdominal cross-sectional imaging techniques [1]. In asymptomatic subjects, the prevalence of pancreatic cysts on abdominal imaging ranges from 2% to 16%, and increases with advancing age [2,3]. Various pathological entities of pancreas may present with the radiological diagnosis of cystic lesions, including benign, borderline, and malignant neoplasms, and non-neoplastic pancreatic cysts [3]. The common cystic neoplasms which are considered benign include serous cystadenomas and pseudocysts, whereas mucinous cystadenomas and intraductal papillary mucinous neoplasms (IPMN) are the common potentially malignant or malignant lesions that require careful analysis [4].
Differential diagnosis is clinically important in order to allow proper management of serous cystadenomas which are benign and surgery should be avoided or minimized, and mucinous cystadenomas which are potential malignant and deserve surgical resection [5,6]. Patient demographics, high-quality cross-sectional imaging, endoscopic ultrasound (EUS) and cyst fluid analysis have been reported to be useful in the differential diagnosis of pancreatic cystic neoplasms [6,7]. However, the accuracy of preoperative diagnosis is still relative low, ranging from 47% to 78% [8-11]. Many of these lesions remain difficult to classify without operative resection.
Computed tomography (CT) is most widely used in the visualization and differentiation of pancreatic cysts based on morphological features, such as location, size, contour, calcifications of cyst wall, septa, and mural nodules [12,13]. However, the accuracy of these morphological characteristic in the differential diagnosis is still unsatisfactory. In the past years, interest has grown in computerized texture analysis of medical images that provides a more detailed and reproducible quantitative assessment of cancer lesion characteristics. Texture analysis refers to a number of mathematical methods that can be used to describe the intensities and spatial distributions of pixels [14]. Texture analysis has been reported to be a reliable technique for differential diagnosis of benign and malignant neoplasms of breast and thyroid [15,16]. However, for the discrimination of pancreatic serous cystadenomas and mucinous cystadenomas, few applications of texture analysis of medical images have been reported. In this research, we assessed the diagnostic role of textural features, and evaluated the combined performance of morphological and textural features in the differential diagnosis of pancreatic serous cystadenomas and mucinous cystadenomas.

Patient population
The Ethics Administration Office of West China Hospital, Sichuan University approved this retrospective study and waived the requirement for informed consent. Patients who were histopathological diagnosed with pancreatic serous or mucinous cystadenoma at our institution between January 2011 and October 2018 were identified from electronic database. Patients without preoperative contrast-enhanced CT images were excluded.
Thirty-two patients with mucinous cystadenomas and 59 patients with serous cystadenomas were enrolled. The selection process of patients was shown in the Additional Figure 1.

Image acquisition and texture analysis
All patients underwent contrast-enhanced CT examination of abdomen following injection of 1.5-2.0 mL/kg of an anionic contrast medium (Omnipaque 350, GE Healthcare) at a rate of 3 mL/s. The images were obtained at a 5 mm section thickness after a 60-65 second delay, with the following acquisition parameters: 120 kVp; 200 to 250 mAs; pitch, 0.75-1.5; collimation, 0.625 mm. All CT examinations were performed using one of the scanners: Brilliance-64, Philips Medical Systems, Eindhoven, The Netherlands; 128-MDCT scanner Somatom Definition, Siemens Healthcare Sector, Forchheim, Germany. Texture analysis of the contrast-enhanced CT images was performed using LifeX software (http://www.lifexsoft.org) [17]. A three-dimensional region of interest (ROI) around the margin of lesion was drawn manually by an experienced abdominal radiologist and textural parameters were retrieved from the ROI. The following 6 groups of textural indices were extracted: histogram, shape and size, gray-level co-occurrence matrix (GLCM), neighborhood gray-level different matrix (NGLDM), gray level run length matrix (GLRLM), and gray-level zone-length matrix (GLZLM).

Statistical Analysis
The least absolute shrinkage and selection operator (LASSO) method was applied to select the textural features. All textural data were given as mean ± standard deviation.
Statistical differences of textural parameters of the patients were analyzed using the Manne-Whitney U test. A p value of less than 0.05 was considered to indicate statistical significance. Receiver operating characteristic curve (ROC) analysis was conducted to estimate the performance of textural features, morphological characteristics, and their combination in the differential diagnosis of serous cystadenomas and mucinous cystadenomas, with the area under the receiver operating characteristic curve (AUC) as the main indicator. Diagnostic accuracy based on the AUC value is defined as follows: 0.9-1.0, excellent; 0.8-0.9, good; 0.7-0.8, moderate; 0.6-0.7, fair; and 0.5-0.6, poor [18]. All statistical analyses were performed using PYTHON software and SPSS version 20.0 (IBM Corporation, Armonk, NY, USA).

Patient population
Baseline characteristics of the patients were summarized in Table 1

Receiver operating characteristic analysis
To discriminate between pancreatic mucinous cystadenomas and serous cystadenomas groups, the AUC of textural parameter with statistical significance between mucinous and serous cystadenomas groups were calculated. The results of ROC analysis were shown in Table 3 and Figure 2. The AUC of SHAPE_Volume (mL), GLRLM_SRHGE, GLRLM_GLNU and GLZLM_GLNU were larger than or equal to 0.700, which were 0.700 (95% confdence

Discussion
Mucinous cystadenomas constitute approximately 23% of all the resected cystic lesions of pancreas, and serous cystadenomas account for 16% [19]. Mucinous cystadenomas have considerable malignant potential, with estimates ranging from 10% to 50% [20]. In contrast, serous cystadenomas are considered benign and are typically found incidentally.
A large multicenter study found only 3 cases of serous adenocarcinomas in a series of 2622 patients with serous cystadenoma, which suggested that serous cystadenomas are almost always benign and indolent tumor [21]. Thus, surgical intervention should be proposed in a minority of patients with serous cystadenoma, only for those who had uncertain diagnosis after systemic examinations or had symptoms [21,22]. Given the risk of invasive disease and the relatively young age at diagnosis, surgical management is recommended for all mucinous cystadenomas patients who are medically fit for the surgery [23]. Therefore, the differential diagnosis of the two diseases is clinically crucial to the treatment regimen options.
Although CT images enable the correct diagnosis in typical cases, serous cystadenomas, especially macrocystic or oligocystic type, is difficult to distinguish from mucinous cystadenomas [24]. Previous studies have reported many cases of pancreatic serous cystadenoma that are misdiagnosed as mucinous cystadenoma and therefore are inappropriately managed [24][25][26]. In this study, the results showed that morphological features and textural parameters including location, size, lobulated contour, Texture analysis refers to a variety of mathematical methods that could be used to describe the position and intensity of signal features, which provides a useful way to maximize the information that can be derived from medical images [14]. Many previous studies focused on textural features have been performed. It has been proposed that textural parameters extracted from the disease lesions can be used to discriminate benign and malignant breast tumors, benign and malignant thyroid nodules, pancreatic lymphoma and pancreatic adenocarcinoma, as well as primary and metastatic lung lesions [15,16,30,31]. However, less attention being paid to textural features of pancreatic cystadenomas, which may be helpful in discrimination of serous and mucinous cystadenomas. In the present study, the results demonstrated that textural parameters were relative good indices in the differentiation of serous and mucinous cystadenomas.
Furthermore, the combination of morphological and texture analysis can significantly improve the diagnostic performance. As an AUC > 0.8 indicated a good accuracy, this combination is considered to be able to distinguish between pancreatic mucinous cystadenomas and serous cystadenomas, and it has potential clinical practical value [18].
There are several limitations in this study. Firstly, the number of patients is relatively small. Second, this is a retrospective analysis in a single center. Third, there is subjectivity in the process of manually outlining the lesion boundary. Therefore, prospective studies with a large population are required to confirm the validity of the present findings.

Conclusions
In conclusion, our preliminary results highlighted the potential of CT texture analysis to discriminate pancreatic serous cystadenomas and mucinous cystadenomas. Furthermore, the combination of morphological features and texture analysis can significantly improve differential diagnostic performance, which may provide a reliable method for selecting pancreatic cystadenoma patients who need surgical intervention.

Ethics approval and consent to participate
The Ethics Administration Office of West China Hospital, Sichuan University approved this retrospective study and waived the requirement for informed consent.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing Interests
The authors declare that they have no competing interests

Authors' contributions
JY designed the study, performed the data analysis and drafted the manuscript. XG and JS performed the data analysis and drafted the manuscript. WZ extracted the data. XM designed the study. All authors read and approved the final manuscript.  Heat map of the textural features and morphological characteristics for differentiating between pancreatic mucinous cystadenoma and serous cystadenoma.