Identification of pancreatic nonfunctional neuroendocrine tumors and solid pseudopapillary tumors via the construction of a consensus clustering model based on enhanced CT images

Identification of pancreatic nonfunctional neuroendocrine tumors and solid pseudopapillary tumors via the construction of a consensus clustering model based on enhanced CT images | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Identification of pancreatic nonfunctional neuroendocrine tumors and solid pseudopapillary tumors via the construction of a consensus clustering model based on enhanced CT images Wuyang Zhang, Peng Cheng, Wei Cao, Min Wang, Bin Wang, Dan Shi, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5782491/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background : Unsupervised clustering has played a greater role in the diagnosis and differential diagnosis of pancreatic tumors in recent years. This study aimed to investigate the value of constructing a c [1] lustering model for unsupervised learning based on enhanced CT to identify pancreatic nonfunctional neuroendocrine tumors (NF-pNETs) and solid pseudopapillary tumors (SPTs). Methods : 45 patients with SPTs and 47 patients with NF-psNETs were retrospectively analyzed. The data were randomly divided into a training set and a validation set at a ratio of 7:3. One-way logistic regression was performed for each clinical variable to assess its relationship with the clustered labels, and a logistic regression model was fitted with the clustered labels as the dependent variable and the clinical variables as independent variables. Variables with P values <0.1 were selected for further analysis. Multifactorial logistic regression models were fitted via the clinical variables selected in the univariate analysis, ridge regularization (L2 penalty) was used to prevent overfitting and address potential multicollinearity, with the strength of regularization (alpha) set to a default value of 1. Clinical variables that were meaningful in the multifactorial logistic regression analyses were used to construct the final logistic regression model, which was used to predict individual clinical-based group labeling on the basis of individual clinical characteristics. For imaging, the optimal number of clusters k selected by the covariance coefficient was used to build the clustering model via unsupervised classification of the lesions through consensus clustering analysis fusing the imaging histological features of arterial-phase, venous-phase, and delayed-phase images. The clinical factors with a final p value < 0.2 were subsequently combined with the clustering model to perform stepwise multivariate logistic regression analyses, thereby establishing a joint model. The performance of the three models was assessed via AUC values, and column line plots were generated to visualize the models. Finally, the clinical validity of the models was assessed via decision curve analysis (DCA). Results : The AUCs of the clustered and clinical models were 0.70 and 0.74 (95% CI: 0.66–0.81) in the training set and 0.65 and 0.75 (95% CI: 0.64–0.87) in the validation set, and the AUCs of the training and test sets in the joint model were 0.88 (95% CI: 0.81–0.94) and 0.85 (95% CI: 0.71–0.96), respectively. Decision curve analysis revealed that the joint clinical-clustering model had a greater net benefit when the high-risk threshold probability was in the range of 0--1. Conclusions : Unsupervised clustering models based on enhanced CT have potential for discriminating between SPTs and NF-PNETs, which can inform clinical decision-making. Imaging histology Unsupervised learning Consensus clustering Pancreatic nonfunctional neuroendocrine tumor Pancreatic solid pseudopapillary tumor Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background Pancreatic neuroendocrine tumors (pNETs) originate from pancreatic pluripotent neural stem cells, are rare among pancreatic tumors, accounting for approximately 1%-5% of all pancreatic tumors[ 1 , 2 ], and are classified into functional and nonfunctional pNETs according to their presence or absence of clinical manifestations[ 3 ]. Pancreatic solid pseudopapillary tumors (SPTs) are also rare tumors of the pancreas, accounting for 2%-3% of all pancreatic exocrine tumors[ 3 ]. NF-pNETs have shown poor long-term survival rates in recent years[ 4 ]. unctional pNETs are easier to distinguish from SPTs because of specific clinical symptoms caused by abnormal hormone secretion. However, nonfunctional pNETs are difficult to differentiate from SPTs because of the lack of specific clinical symptoms. At present, enhanced CT is a commonly used imaging method for identifying pancreatic tumors and is an important auxiliary tool for clinicians in the diagnosis and differential diagnosis of pancreatic masses[ 5 – 7 ]. Nevertheless, the capacity of enhanced CT to accurately distinguish pNETs from SPTs remains imperfect. The definitive diagnosis ultimately hinges on a pathological assessment, which inherently introduces greater complexity and discomfort for patients. The necessity for a noninvasive quantitative approach to differentiate between the two is therefore paramount. A clear preoperative diagnosis is highly valuable in the selection of treatment modalities and prognosis prediction for patients. Radiomics is an emerging field of research because medical images reflect underlying pathophysiological features, high-throughput extraction of features from imaging images in a noninvasive manner, analysis and interpretation of quantitative imaging parameters, selection of highly relevant features after dimensionality reduction processing, and combination of these features to construct predictive or representational models[ 8 , 9 ]. Machine learning is classified into supervised and unsupervised learning according to the model type. Supervised learning usually uses labeled training data for modeling, which is usually time-consuming and labor-intensive, whereas unsupervised learning does not require labeled data for modeling. Traditional imaging genomics is usually computed via supervised machine learning, which generally requires large amounts of labeled data, which is time-consuming and labor-intensive for clinicians. Unsupervised learning has the potential to elucidate the biological processes underlying similar or dissimilar imaging phenotypes and clinical manifestations of tumors because of its ability to process large-scale, high-dimensional datasets without the need for prior labeling or guidance and to discover potential heterogeneity within them[ 10 – 12 ]. The most common unsupervised learning method is cluster analysis, and common clustering methods include k-means clustering, hierarchical clustering, and mean drift clustering. Consensus cluster analysis[ 13 ] is a technique that combines multiple clusters into a more stable cluster. This is achieved by first extracting a certain sample of subdatasets via a resampling method and then dividing each subsample into a maximum of k groups via a specified clustering algorithm. Finally, the results of cluster analysis are combined after multiple resamplings to yield a consistent assessment. As a consequence of the enhanced stability of the resulting clusters with respect to the sampling variance, it can be posited that this consistent clustering outcome represents a genuine subclass. Consequently, consensus clustering mitigates the risk of model overfitting, enhances the robustness of cluster groups, and facilitates independent access to individual partitions while also facilitating more effective handling of missing values. Accordingly, this study constructed a consensus clustering model to distinguish NF-pNETs and SPTs on the basis of enhanced CT images, with the objective of developing a more effective preoperative differential diagnosis method for clinical use. Methods Study population Imaging and preoperative clinical data of patients who underwent surgical resection of the pancreas from January 2016 to December 2022 and were pathologically confirmed to have solid pseudopapillary tumors of the pancreas (n = 45) and neuroendocrine tumors of the pancreas (n = 47) were retrospectively analyzed (Fig. 1 ). All patients underwent enhanced CT within 15 days before surgery. The inclusion criterion was that both NF-pNETs and SPTs were confirmed by pathology. The exclusion criteria were (1) receiving radiotherapy or chemotherapy before CT examination and (2) poor-quality preoperative CT images. This was a retrospective study approved by the Ethical Review Committee of the First Hospital of the University of Science and Technology of China, which agreed to waive informed consent. Imaging and clinical data Three-phase enhancement images of the lesion, sex, age, CA199, CA125, CEA, BMI, abdominal pain, diabetes mellitus, hypertension, tumor site, metastasis, tumor size, morphology, boundary, texture, calcification, biliary dilatation, and lymph nodes, etc. Method of examination A GE Discovery CT 750 multislice CT scanner was used to scan from the diaphragmatic apex to the level of the iliac spine. The scanning parameters were as follows: tube voltage of 120 kVp, tube current of 240 ~ 300 mA, layer thickness of 5 mm, reconstruction of 1.25 mm, layer spacing of 5 mm, and pitch of 1.375 : 1. The patient underwent CT scanning of the abdomen, followed by administration of the nonionic contrast agent iodohexol (300 mgI/ml) via the elbow vein mass at a dose of 1.5 ml/kg at a rate of 3.0 ml/s; the following steps were carried out in the sequence of the arterial (30 s), portal venous (70 s), and delayed (180 ~ 300 s) phases of the scan. Lesion segmentation and feature extraction Preoperative enhanced CT images of the patient were collected and exported from the PACS workstation in DICOM format, and the axial continuous level images of the patient's three preoperative enhanced CT phases were selected for manual ROI outlining by two senior physicians who had been engaged in diagnostic CT imaging for more than 15 years. ITK-SNAP software (Fig. 2 ) was used, and the two radiologists, both of whom were unaware of the histopathological results, avoided blood vessels during the outlining process. Calcification and other components were avoided during the outlining process, and when the outlining results were inconsistent, both radiologists made a joint decision via consultation. Two weeks later, the interclass and intraclass correlation efficiency (ICC) were calculated separately to assess observer agreement. Thirty patients were randomly selected from all patients with preoperative enhanced CT images, one of the doctors resketched the lesions to assess intraclass agreement, and the other doctor similarly sketched the same 30 lesions to assess interclass agreement. concordance, and the features with an ICC ≥ 0.75 were then analyzed in the next step. Feature extraction was performed via Image Standardization Initiative (IBSI)-compliant AK (3.3.0, GE Healthcare) software, and the types of features extracted included first-order statistical features, morphological features, grayscale covariance matrix features, grayscale tour features, and wavelet-variation and Laplace-variation features. Unsupervised consensus cluster analysis To ensure data integrity and consistency, the clinical dataset was meticulously preprocessed. Missing values, outliers and irregular data formats were corrected and adjusted. Categorical variables were processed via either solo thermal coding or ordered coding according to their characteristics, enabling them to be adapted to subsequent analytical models. Continuous variables, on the other hand, were standardized to ensure that they had a mean of 0 and a standard deviation of 1. To dig deeper into the internal structure and potential subgroups of the data, an unsupervised consensus clustering method was employed. Consensus clustering is an advanced technique for assessing the stability of clustering results. It provides important information about the stability of the clustering structure by clustering the dataset multiple times and observing the common belonging of samples in different clusters. The core element is that if two samples are frequently attributed to the same cluster over multiple clustering iterations, the consensus between them is greater. This relationship is explicitly described by generating a consensus matrix, where each element represents how often two samples are assigned to the same cluster across all clustering iterations. In addition, this matrix can be further clustered to obtain a robust cluster partitioning of the dataset. To find the optimal number of clusters (k), we use the covariance coefficient to evaluate the fitness of different k values. The covariance coefficient is a measure of the strength of the relationship between two variables that provides us with information about the stability of the cluster structure at each k value. Theoretically, a higher value of the coefficient of the symbiotic relationship indicates that the k value provides a better clustering structure for the dataset. By comparing the coefficients of symbiotic relationships at different values of k, we can determine an optimal value of k that maximizes that coefficient. Statistical analyses and modeling All the statistical analyses were performed in Python (version 3.8.2, http://www.python.org ). No statistical models were developed, and univariate logistic regression was performed for each clinical variable to assess its relationship with the clustered labels, fitting a logistic regression model with the clustered labels as the dependent variable and the clinical variables as the independent variables. Variables with P values less than 0.1 were selected for further analysis. Multivariate logistic regression models were fitted using the clinical variables and consensus clustering categories selected in the univariate analyses, Ridge regularization (L2 penalty) was used to prevent overfitting and address potential multicollinearity, with the strength of regularization (alpha) set to the default value of 1. The final logistic regression was constructed using the statistically significant clinical variables identified in the multivariate logistic regression analysis model, which was used to predict group labeling on the basis of individual clinical characteristics. The performance of the model was assessed via metrics such as the area under the receiver operating characteristic (ROC) curve. The coefficients of the final logistic regression model were interpreted for relationships between clinical variables, clusters and NF-pNETs/SPTs, and the predictive performance of the model is reported. The final p < 0.2 clinical factors were then combined with the cluster model in a stepwise multivariate logistic regression analysis, leading to a joint model and presentation of the model in a column-line diagram. The predictive ability of the three models was evaluated and compared via the area under the curve (AUC) of the ROC curve as well as model accuracy, sensitivity, specificity and negative and positive predictive values (Fig. 3 ). Results Study population From January 2016 to December 2022, 47 SPENTs and 45 SPTs were eligible for this retrospective study and were divided into a training set consisting of 17 males and 47 females (range 9–76 years, mean 42.44 years) and a validation set consisting of 13 males and 15 females (range 11–76 years, mean 46.89 years). Clinical models A subset of clinical characteristics with strong covariates was removed. According to the results of the univariate analysis, the differences in hypertension, CEA, tumor boundary, tumor morphology, tumor texture and enhancement III were statistically significant (P < 0.1), as shown in Table 1 . CEA, tumor boundary, tumor texture, arterial phase and delayed phase CT scans were shown to be clinically independent risk factors for disease development in the results of multifactorial logistic regression analysis (P < 0.2), as shown in Table 2 . Table 1 Comparison of general information. Univariate Regression Analysis(P |z| [0.025 0.975] Hypertension 1.401 0.716 1.956 0.050 -0.002 2.804 Diabetes 9.338 51.561 0.181 0.856 -91.720 110.396 Abdominal Pain -0.366 0.506 -0.724 0.469 -1.358 0.626 BMI -1.968 22.702 -0.087 0.931 -46.464 42.528 CEA 0.933 0.334 2.790 0.005 0.278 1.588 CA199 0.050 0.040 1.223 0.221 -0.030 0.129 CA125 0.011 0.015 0.713 0.476 -0.019 0.040 Tumor Size -0.077 0.074 -1.046 0.296 -0.222 0.068 Tumor Boundary 2.092 0.640 3.270 0.001 0.838 3.346 Tumor Morphology 1.831 0.601 3.049 0.002 0.654 3.008 Tumor Texture -0.804 0.392 -2.051 0.040 -1.572 -0.036 Biliary Dilatation 0.597 0.684 0.873 0.383 -0.744 1.939 Ealcification -0.680 0.549 -1.239 0.215 -1.755 0.395 Lymph Nodes 10.487 192.381 0.055 0.957 -366.572 387.546 Consensus_Group_A -0.346 0.144 -2.397 0.017 -0.629 -0.063 Consensus_Group_D -0.222 0.127 -1.750 0.080 -0.472 0.027 Consensus_Group_V 0.053 0.115 0.461 0.645 -0.172 0.278 Table 2 Multifactor logistic regression analysis. Multifactor regression analysis (P |z| [0.025 0.975] Const 1.077 1.277 0.843 0.399 -1.426 3.579 Hypertension 0.627 0.917 0.684 0.494 -1.171 2.424 CEA 0.672 0.365 1.840 0.066 -0.044 1.387 Tumor Boundary 1.769 0.971 1.822 0.068 -0.134 3.671 Tumor Morphology 0.636 0.871 0.730 0.465 -1.071 2.344 Tumor Texture -0.987 0.506 -1.951 0.051 -1.978 0.005 Consensus_Group_A -0.503 0.226 -2.227 0.026 -0.946 -0.060 Consensus_Group_D -0.215 0.167 -1.291 0.197 -0.542 0.111 Unsupervised consensus clustering model Unsupervised classification of lesions was performed via consensus clustering analysis incorporating the histological features of A, D, and V images, and the optimal numbers of clusters selected by the covariance coefficient, k, were 8, 4, and 7, respectively. The consensus clustering results were plotted according to the clustering results (Fig. 4 A). After one-way logistic regression analysis, the A and D consensus clustering categories had a significant relationship with outcome classification (P values = 0.017 and 0.08, Table 1 ). Unsupervised cluster analysis for neuroendocrine tumors and solid pseudopapillary tumors was performed via a multifactorial logistic regression model incorporating A and D. Joint Clinical-Consensus Clustering Models The most valuable clinical factors and unsupervised cluster analyses were used to establish a joint clinical cluster model via a multifactor logistic regression model. The predictive ability of the above three models was evaluated via model accuracy, sensitivity, specificity, and negative and positive predictive values (Table 3 ), and the predictive ability of the above three models was compared via the area under the ROC curve (Fig. 4 B). The AUC of the training centralized clinical model = 0.81; the AUC increased to 0.88 after adding the cluster analysis results to build the joint model (P < 0.05, Delong test). Validation of the centralized clinical model yielded an AUC = 0.66; after adding the results of cluster analysis to create a joint model, the AUC significantly increased to 0.85 (p < 0.05, Delong test). This shows that the joint model yielded higher values than either the clinical model or the unsupervised clustering model did, which suggests that the joint model features play a highly significant role in discriminating between NF-pNETs and SPTs. The final generated model line diagrams and Noetherian diagrams are shown in Fig. 4 C and Fig. 5 A, B. Finally, we assessed the clinical validity of the above models via DCA (Fig. 5 C), which revealed that the joint clinical-clustering model of unsupervised learning had a greater net benefit when the probability of a high-risk threshold was in the range of 0–1, suggesting that the model had greater clinical validity than each of the other models. Table 3 Performance comparison of different models. AUC [95% CI] ACC [95% CI] SEN [95% CI] SPE [95% CI] PPV [95% CI] NPV [95% CI] Clinical Training 0.81 [0.72–0.90] 0.781 [0.703–0.859] 0.788 [0.676–0.900] 0.774 [0.636–0.893] 0.788 [0.658–0.900] 0.774 [0.654–0.893] Testing 0.66 [0.48–0.83] 0.679 [0.500–0.821] 0.857 [0.692–1.000] 0.500 [0.273–0.727] 0.632 [0.429–0.812] 0.778 [0.500–1.000] Consensus Groups Training 0.70 [0.59–0.80] 0.656 [0.562–0.750] 0.545 [0.400–0.690] 0.774 [0.647–0.889] 0.720 [0.560–0.864] 0.615 [0.486–0.744] Testing 0.65 [0.47–0.82] 0.679 [0.536–0.821] 1.000 [1.000–1.000] 0.357 [0.154–0.583] 0.609 [0.429–0.773] 1.000 [1.000–1.000] Nomogram Training 0.88 [0.81–0.94] 0.812 [0.734–0.891] 0.848 [0.741–0.946] 0.774 [0.649–0.886] 0.800 [0.677–0.903] 0.828 [0.708–0.941] Testing 0.85 [0.71–0.96] 0.786 [0.643–0.893] 0.714 [0.500–0.917] 0.857 [0.692–1.000] 0.833 [0.615–1.000] 0.750 [0.562–0.929] Discussion SPTs and pNETs are both rare tumors of the pancreas, and the detection rate of both has gradually increased in recent years with the rapid development of medical imaging. The identification of NF-pNETs is more difficult than that of functional pNETs because of the absence of specific clinical symptoms. NF-pNETs are more aggressive because of their more invasive nature and more aggressive surgical approaches, including pancreatectomy and regional lymph node dissection, thereby achieving negative tumor margins[ 14 , 15 ]. Therefore, accurate preoperative diagnosis is important for the selection and guidance of clinical treatment. Currently, early screening of pancreatic tumors relies on imaging, and both NFPNETs and SPTs can present as well-defined round or round-like masses with solid, cystic, or cystic-solid borders, and enhancement can show insignificant or progressive enhancement[ 5 , 6 ]. It is not always possible to accurately distinguish between the two on conventional imaging. Moreover, statistically significant differences in the clinical characteristics of patients with NF-pNETs and those with SPTs in terms of sex, age, calcification and tumor size have been reported in the literature, with NF-pNETs mostly observed in middle-aged and elderly people and SPTs being more common in young women[ 16 , 17 ]; however, it is more difficult to differentiate between atypical NF-PNETs and SPTs. Therefore, in this study, along with the basis of clinical factors combined with imaging histology, quantitative imaging histology cluster analysis will help to differentiate between the two and help clinicians plan surgery and choose combined treatment options. In this study, the clinical characteristics with strong covariates were first removed (sex and age) in the analyses targeting the clinical factors, indicating that these two factors differed significantly between SPTs and NF-PNETs. This difference occurred mainly in young women with a mean age of 34.3 years and in NF-PNETs, who were mainly in the middle-aged and older age groups with a mean age of 52.1 years, which is in line with the majority of the results in the literature. In addition, in the multifactorial analysis of clinical and imaging characteristics in this study, the CEA level, tumor border, tumor texture, presence of arterial-stage lesions and presence of delayed-stage lesions were found to be independent risk factors for differentiating NF-PNETs from SPTs. Most SPTs are relatively inert tumors with well-defined borders and intact perimeters, whereas NF-PNETs have no specific clinical manifestations, are found with larger lesions, are more likely to develop into malignant tumors, and exhibit invasive features such as blurred borders and incomplete perimeters, which is similar to the results of this study. The texture can affect the occurrence of both diseases. From the pathological point of view, solid pseudopapillary tumor tumor cells degenerate and shed to form cystic areas, the size of the cystic cavity varies, and the surrounding tumor cells are arranged in the form of trabecular or glandular vesicles. The tumor in the solid area is in the form of a sheet of lumps or diffusely distributed, and it is separated by small, fibrous blood vessels. The pseudopapillary area can be seen in the tumor cells surrounded by the surrounding fibrous blood vessels in the separation of the complex arrangement of forming a characteristic pseudo-pseudopapillary area, in which the tumor cells are surrounded by surrounding fibrous vessels to form a characteristic pseudopapillary structure[ 18 ]; NF-pNETs have a variety of histological structures under the microscope, and the cells can be arranged into solid nest-like, trabecular, gyrus-like, and glandular vesicles, etc., and may show cystic changes in the event of necrosis[ 19 ]. The SPTs in this study were mainly cystic-solid and cystic, and the NF-PNETs were mainly solid, which is consistent with the pathological manifestations of these two types of tumors. The CT manifestations were based on the density differences presented by the pathological manifestations. In the present study, the enhancement of the arterial and delayed phases on enhanced CT was found to be meaningful in distinguishing SPTs from NF-PNETs. It is possible that NF-pNETs contain more abundant blood vessels and that the enhancement in the arterial phase is significant. SPT tumor cells are arranged in nested sheets and divided by small blood vessels into blood sinusoidal structures similar to those of cavernous hemangiomas, so the enhancement scans show progressive enhancement. In addition, CEA was statistically significant in this study, but its statistical significance was not mentioned in the relevant literature, which may be caused by the small sample size of this study. Currently, to circumvent the limitations of clinical features and imaging features, quantitative research methods such as imaging methods and imaging histology are carried out in the clinic. Most of the previous literature has studied the imaging features of SPTs and NF-PNETs, but this is too one-sided for atypical patients. In this study, on the one hand, we established an unsupervised clustering model, and the AUC values of the model in the training set and validation set were 0.70 (95% CI: 0.59–0.80) and 0.65 (95% CI: 0.47–0.82), respectively, with good predictive efficacy, which suggests that quantitative imaging of histological features can be used as a new way to identify SPTs and NFPNETs. On the other hand, when we combined the clinical factors and cluster analysis, the predictive efficacy was more significantly improved, with AUC values of 0.88 (95% CI: 0.81–0.94) and 0.85 (95% CI: 0.71–0.96) for the training and validation sets, respectively, suggesting that clinical factors are also particularly important for both tumors and that a combination of the two can maximize the benefits to patients. The combined clinical-clustering model in this study also provides a new method and idea for diagnosing SPTs and NF-PNETs. In addition, the clustering model developed in this study uses an unsupervised machine learning approach. Unsupervised consensus clustering is an advanced clustering technique for unsupervised learning. Unlike traditional one-shot clustering methods, this method clusters the data multiple times and in multiple ways and observes the common attributes of the samples across the clustering results. The main advantage of this method is its robustness to results, which significantly reduces biases introduced by random initializations or specific algorithms. In addition, it is robust and adaptable for handling a wide range of data types and structures, helping to account for subtle data structures that may have been overlooked in word clustering. Moreover, evaluating the consensus under different numbers of clusters can provide a strong guide for determining the optimal number of clusters. The limitations of this study are as follows: (i) The sample size is small, which may lead to model overfitting and have some impact on the stability of the model. (ii) Some disease-related laboratory indicators, such as neuron-specific enolase and chromogranin A, were not included. (iii) The evaluation and interpretation of unsupervised machine learning models may be relatively difficult due to the lack of labeling information. In addition, unsupervised learning algorithms usually require more computational resources and more complex algorithm designs to discover patterns efficiently from data. In summary, this study constructed an unsupervised clustering model based on enhanced CT combined with clinical features to construct a comprehensive model to distinguish between NF-pNETs and SPTs, which has great potential for identifying patients requiring monitoring or surgical resection. Conclusions Our study found that unsupervised clustering models based on augmented CT potentially differentiate between SPTs and NF-PNETs, which can provide a new diagnostic basis for diagnostic imaging in the future. Abbreviations Abbreviation Annotation ACC Accuracy AK Artificial Intelligence Kit AUC Area Under the Curve BMI Body Mass Index CA125 Cancer Antigen 125 CA199 Carbohydrate Antigen 19-9 CEA Carcinoembryonic Antigen CI Confidence Interval coef Coefficient CT Computed Tomography DICOM Digital Imaging and Communications in Medicine GE General Electric ICC Intraclass Correlation Coefficient IBSI Image Biomarker Standardization Initiative NF-pNETs Non-Functioning Pancreatic Neuroendocrine Tumors NPV Negative Predictive Value PACS Picture Archiving and Communication Systems pNETs Pancreatic Neuroendocrine Tumors PPV Positive Predictive Value SEN Sensitivity SPTs Solid Pseudopapillary Tumors SPE Specificity std err Standard Error Declarations Ethics approval and consent to participate This study has been approved by the Institutional Ethics Review Committee of the First Affiliated Hospital of the University of Science and Technology of China (No. 2024-RE-464), and the Ethics Review Committee of the First Affiliated Hospital of the University of Science and Technology of China has waived informed individual consent for this retrospective analysis. All methods are executed in accordance with relevant regulations and guidelines. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Authors details 1 Graduate Department, Bengbu Medical University, Bengbu, Anhui, China 2 Department of Radiology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Sciences and Technology of China, Hefei 230001, China 3 Graduate Department, Wannan Medical College, Wuhu, Anhui, China Funding This study was supported by grants from the National Natural Science Foundation of China (82271991) and the Anhui Province Key R&D Life and Health Category A Project (2022e0702007). No other potential conflicts of interest relevant to this article are reported. Author Contribution WZ and WW were responsible for research design and project management. PC was responsible for data collection. WC and MW were responsible for image processing and feature extraction. BW was responsible for article writing. DS and JL were responsible for statistical analysis. All authors contributed to the article and approved the submitted version. Availability of data and materials The dataset generated for this study can be obtained from the correspondence author. References Wang Y, Miller FH, Chen ZE, Merrick L, Mortele KJ, Hoff FL, Hammond NA, Yaghmai V, Nikolaidis PJR: Diffusion-weighted MR imaging of solid and cystic lesions of the pancreas . 2011, 31 (3):E47-E64. Low G, Panu A, Millo N, Leen EJR: Multimodality imaging of neoplastic and nonneoplastic solid lesions of the pancreas . 2011, 31 (4):993-1015. Niederle B, Selberherr A, Bartsch DK, Brandi ML, Doherty GM, Falconi M, Goudet P, Halfdanarson TR, Ito T, Jensen RTJN: Multiple endocrine neoplasia type 1 and the pancreas: diagnosis and treatment of functioning and non-functioning pancreatic and duodenal neuroendocrine neoplasia within the MEN1 syndrome–an international consensus statement . 2021, 111 (7):609-630. Wu J, Sun C, Li E, Wang J, He X, Yuan R, Yi C, Liao W, Wu LJBc: Non-functional pancreatic neuroendocrine tumours: emerging trends in incidence and mortality . 2019, 19 :1-10. Brook OR, Gourtsoyianni S, Brook A, Siewert B, Kent T, Raptopoulos VJR: Split-bolus spectral multidetector CT of the pancreas: assessment of radiation dose and tumor conspicuity . 2013, 269 (1):139-148. Marin D, Nelson RC, Barnhart H, Schindera ST, Ho LM, Jaffe TA, Yoshizumi TT, Youngblood R, Samei EJR: Detection of pancreatic tumors, image quality, and radiation dose during the pancreatic parenchymal phase: effect of a low-tube-voltage, high-tube-current CT technique—preliminary results . 2010, 256 (2):450-459. Scialpi M, Reginelli A, D'Andrea A, Gravante S, Falcone G, Baccari P, Manganaro L, Palumbo B, Cappabianca SJIjos: Pancreatic tumors imaging: An update . 2016, 28 :S142-S155. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, Van Stiphout RG, Granton P, Zegers CM, Gillies R, Boellard R, Dekker AJEjoc: Radiomics: extracting more information from medical images using advanced feature analysis . 2012, 48 (4):441-446. Mukherjee S, Patra A, Khasawneh H, Korfiatis P, Rajamohan N, Suman G, Majumder S, Panda A, Johnson MP, Larson NBJG: Radiomics-based machine-learning models can detect pancreatic cancer on prediagnostic computed tomography scans at a substantial lead time before clinical diagnosis . 2022, 163 (5):1435-1446. e1433. Bröker F, Holt LL, Roads BD, Dayan P, Love BCJTiCS: Demystifying unsupervised learning: how it helps and hurts . 2024. Tobaly D, Santinha J, Sartoris R, Dioguardi Burgio M, Matos C, Cros J, Couvelard A, Rebours V, Sauvanet A, Ronot MJC: CT-based radiomics analysis to predict malignancy in patients with intraductal papillary mucinous neoplasm (IPMN) of the pancreas . 2020, 12 (11):3089. Wan Y, Yang P, Xu L, Yang J, Luo C, Wang J, Chen F, Wu Y, Lu Y, Ruan DJMP: Radiomics analysis combining unsupervised learning and handcrafted features: A multiple‐disease study . 2021, 48 (11):7003-7015. Lock EF, Dunson DBJB: Bayesian consensus clustering . 2013, 29 (20):2610-2616. Akirov A, Larouche V, Alshehri S, Asa SL, Ezzat SJC: Treatment options for pancreatic neuroendocrine tumors . 2019, 11 (6):828. Katsoulakis E, Yu Y, Apte AP, Leeman JE, Katabi N, Morris L, Deasy JO, Chan TA, Lee NY, Riaz NJOo: Radiomic analysis identifies tumor subtypes associated with distinct molecular and microenvironmental factors in head and neck squamous cell carcinoma . 2020, 110 :104877. Wang C, Cui W, Wang J, Chen X, Tong H, Wang ZJAR: Differentiation between solid pseudopapillary neoplasm of the pancreas and hypovascular pancreatic neuroendocrine tumors by using computed tomography . 2019, 60 (10):1216-1223. Liu F, Li J, Fang X, Meng Y, Zhang H, Yu J, Feng X, Wang L, Jiang H, Lu JJAR: Differentiation of Solid Pseudopapillary Tumor and Non-Functional Neuroendocrine Tumors of the Pancreas Based on CT Delayed Imaging: A Propensity Score Analysis . 2022, 29 (3):350-357. Hao EIU, Hwang HK, Yoon D-S, Lee WJ, Kang CMJM: Aggressiveness of solid pseudopapillary neoplasm of the pancreas: a literature review and meta-analysis . 2018, 97 (49):e13147. Pathology DoDDJCJo: Chinese consensus on the pathological diagnosis of gastrointestinal and pancreatic neuroendocrine neoplasms (2020)(in Chinese) . 2021, 50 :14-20. Additional Declarations No competing interests reported. Supplementary Files Clinicaldata.xls Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5782491","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":400070194,"identity":"d815ac96-bd3d-4ee0-8a0e-58a6c5a82575","order_by":0,"name":"Wuyang Zhang","email":"","orcid":"","institution":"Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Wuyang","middleName":"","lastName":"Zhang","suffix":""},{"id":400070196,"identity":"ead04f4f-5925-4944-a554-9455bd85d593","order_by":1,"name":"Peng Cheng","email":"","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Peng","middleName":"","lastName":"Cheng","suffix":""},{"id":400070201,"identity":"17f486aa-6a99-4662-880d-1250fd12de0c","order_by":2,"name":"Wei Cao","email":"","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Cao","suffix":""},{"id":400070203,"identity":"2d74b51b-0900-46f4-8594-2e758db79e77","order_by":3,"name":"Min Wang","email":"","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Min","middleName":"","lastName":"Wang","suffix":""},{"id":400070205,"identity":"a09b1237-d08e-417a-99ff-b4566819c96a","order_by":4,"name":"Bin Wang","email":"","orcid":"","institution":"Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Bin","middleName":"","lastName":"Wang","suffix":""},{"id":400070207,"identity":"d4826a7c-414b-4bc9-80cf-6915f74499d6","order_by":5,"name":"Dan Shi","email":"","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Dan","middleName":"","lastName":"Shi","suffix":""},{"id":400070212,"identity":"c062a821-9173-4d9c-9903-70582b02c8c6","order_by":6,"name":"Jingyu Li","email":"","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Jingyu","middleName":"","lastName":"Li","suffix":""},{"id":400070214,"identity":"54a71410-4729-4b4a-8075-847130dd7431","order_by":7,"name":"Wei Wei","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIiWNgGAWjYFACxgYGBgMLGX4kLlFaJHgkGyCqidECBhI8BgeI1WJwvLlN4kOBBI/xjfTnD34w2MhuOMD87AFeLWcOtknOADrM7EaOYWMPQ5rxhgNs5gb4tJjdSGy7zQPRwtjMwHA4ccMBHjYJvFruP2y7/QeoxXhG+kOglv9EaLnB2HYbFGIGEgmGQC0HCGuxP5PY/rMHqEXizBvDmT0GycYzD7OZ4dUi2X78scGPPzZy/O3pDz78qLCT7Tve/AyvFjQACipmEtSPglEwCkbBKMAOAALgSMCOu0LCAAAAAElFTkSuQmCC","orcid":"","institution":"The First Affiliated Hospital of USTC, University of Sciences and Technology of China","correspondingAuthor":true,"prefix":"","firstName":"Wei","middleName":"","lastName":"Wei","suffix":""}],"badges":[],"createdAt":"2025-01-07 15:08:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5782491/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5782491/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73676968,"identity":"36a26497-afa0-443b-80cd-01d2df733671","added_by":"auto","created_at":"2025-01-13 13:13:44","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":84306,"visible":true,"origin":"","legend":"\u003cp\u003eRelevant patient exclusion and inclusion flowchart.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/bd6202743d853fb769776d9a.png"},{"id":73674544,"identity":"460d0e98-e12a-440f-8bcf-01f6313c74f8","added_by":"auto","created_at":"2025-01-13 12:57:44","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1994378,"visible":true,"origin":"","legend":"\u003cp\u003eROI images of two tumor foci. \u003cstrong\u003e(A, B)\u003c/strong\u003e A 51-year-old female with pathology of a (pancreatic tail) neuroendocrine tumor; \u003cstrong\u003e(C, D)\u003c/strong\u003e a 32-year-old female with pathology of a (pancreatic body) solid pseudopapillary tumor.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/0059c45233277eab2f1e98d3.png"},{"id":73674543,"identity":"636492d9-5430-48f1-a122-1420c581fb73","added_by":"auto","created_at":"2025-01-13 12:57:44","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":856040,"visible":true,"origin":"","legend":"\u003cp\u003eStudy process diagram.\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/82120a1a39a9e9e5c0a9c598.png"},{"id":73675426,"identity":"8c9cf454-2a6b-4c87-b564-4ee007a79a65","added_by":"auto","created_at":"2025-01-13 13:05:43","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2468091,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A) \u003c/strong\u003eHeatmap of clustering of arterial/venous/delayed phase images. \u003cstrong\u003e(B)\u003c/strong\u003e ROC curves for the three models: training set in D and test set in E. The nomogram (green) is the federated model; the cluster group (red) is the clustered model; and the clinical (blue) is the clinical model. \u003cstrong\u003e(C)\u003c/strong\u003e Column-linear graphical risk model (nomogram) for the identification of NF-pNETs and SPTs.\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/cc30bc249c71bf0b4bae80f1.png"},{"id":73674534,"identity":"bacc4d04-7dde-4dc3-86aa-9881cf582709","added_by":"auto","created_at":"2025-01-13 12:57:43","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":503314,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A, B) \u003c/strong\u003eCalibration curves for the training and validation sets. The y-axis represents the occurrence of the predicted event rates, and the x-axis represents the actual event rates. The diagonal dashed line represents the ideal prediction of the ideal model, and the blue solid line represents the performance of the nomogram. When it is close to the diagonal line, the predicted probabilities are in good agreement with the true probabilities.\u003cstrong\u003e (C) \u003c/strong\u003eStandardized net benefits of the three models over a range of high-risk thresholds; the higher the standardized net benefit is, the greater the clinical validity of the model. Green represents the joint clinical-clustering model, and blue represents the clinical model alone. Red represents the clustering model alone.\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/f96d8cf2422daa8a12e4614b.png"},{"id":73678902,"identity":"17ed83bc-af0a-474a-b1b6-d217f607a419","added_by":"auto","created_at":"2025-01-13 13:37:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":8463465,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/2e0e6613-607d-4e0a-9082-697db92da01d.pdf"},{"id":73675430,"identity":"f963087a-cc6d-4e63-bfdc-55fc18e8aaff","added_by":"auto","created_at":"2025-01-13 13:05:44","extension":"xls","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":235520,"visible":true,"origin":"","legend":"","description":"","filename":"Clinicaldata.xls","url":"https://assets-eu.researchsquare.com/files/rs-5782491/v1/f7d2efdbfa253c3714ac154a.xls"}],"financialInterests":"No competing interests reported.","formattedTitle":"Identification of pancreatic nonfunctional neuroendocrine tumors and solid pseudopapillary tumors via the construction of a consensus clustering model based on enhanced CT images","fulltext":[{"header":"Background","content":"\u003cp\u003ePancreatic neuroendocrine tumors (pNETs) originate from pancreatic pluripotent neural stem cells, are rare among pancreatic tumors, accounting for approximately 1%-5% of all pancreatic tumors[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], and are classified into functional and nonfunctional pNETs according to their presence or absence of clinical manifestations[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Pancreatic solid pseudopapillary tumors (SPTs) are also rare tumors of the pancreas, accounting for 2%-3% of all pancreatic exocrine tumors[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. NF-pNETs have shown poor long-term survival rates in recent years[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. unctional pNETs are easier to distinguish from SPTs because of specific clinical symptoms caused by abnormal hormone secretion. However, nonfunctional pNETs are difficult to differentiate from SPTs because of the lack of specific clinical symptoms. At present, enhanced CT is a commonly used imaging method for identifying pancreatic tumors and is an important auxiliary tool for clinicians in the diagnosis and differential diagnosis of pancreatic masses[\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Nevertheless, the capacity of enhanced CT to accurately distinguish pNETs from SPTs remains imperfect. The definitive diagnosis ultimately hinges on a pathological assessment, which inherently introduces greater complexity and discomfort for patients. The necessity for a noninvasive quantitative approach to differentiate between the two is therefore paramount.\u003c/p\u003e \u003cp\u003eA clear preoperative diagnosis is highly valuable in the selection of treatment modalities and prognosis prediction for patients. Radiomics is an emerging field of research because medical images reflect underlying pathophysiological features, high-throughput extraction of features from imaging images in a noninvasive manner, analysis and interpretation of quantitative imaging parameters, selection of highly relevant features after dimensionality reduction processing, and combination of these features to construct predictive or representational models[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Machine learning is classified into supervised and unsupervised learning according to the model type. Supervised learning usually uses labeled training data for modeling, which is usually time-consuming and labor-intensive, whereas unsupervised learning does not require labeled data for modeling. Traditional imaging genomics is usually computed via supervised machine learning, which generally requires large amounts of labeled data, which is time-consuming and labor-intensive for clinicians. Unsupervised learning has the potential to elucidate the biological processes underlying similar or dissimilar imaging phenotypes and clinical manifestations of tumors because of its ability to process large-scale, high-dimensional datasets without the need for prior labeling or guidance and to discover potential heterogeneity within them[\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The most common unsupervised learning method is cluster analysis, and common clustering methods include k-means clustering, hierarchical clustering, and mean drift clustering. Consensus cluster analysis[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] is a technique that combines multiple clusters into a more stable cluster. This is achieved by first extracting a certain sample of subdatasets via a resampling method and then dividing each subsample into a maximum of k groups via a specified clustering algorithm. Finally, the results of cluster analysis are combined after multiple resamplings to yield a consistent assessment. As a consequence of the enhanced stability of the resulting clusters with respect to the sampling variance, it can be posited that this consistent clustering outcome represents a genuine subclass. Consequently, consensus clustering mitigates the risk of model overfitting, enhances the robustness of cluster groups, and facilitates independent access to individual partitions while also facilitating more effective handling of missing values. Accordingly, this study constructed a consensus clustering model to distinguish NF-pNETs and SPTs on the basis of enhanced CT images, with the objective of developing a more effective preoperative differential diagnosis method for clinical use.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy population\u003c/h2\u003e \u003cp\u003eImaging and preoperative clinical data of patients who underwent surgical resection of the pancreas from January 2016 to December 2022 and were pathologically confirmed to have solid pseudopapillary tumors of the pancreas (n\u0026thinsp;=\u0026thinsp;45) and neuroendocrine tumors of the pancreas (n\u0026thinsp;=\u0026thinsp;47) were retrospectively analyzed (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). All patients underwent enhanced CT within 15 days before surgery. The inclusion criterion was that both NF-pNETs and SPTs were confirmed by pathology. The exclusion criteria were (1) receiving radiotherapy or chemotherapy before CT examination and (2) poor-quality preoperative CT images. This was a retrospective study approved by the Ethical Review Committee of the First Hospital of the University of Science and Technology of China, which agreed to waive informed consent.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eImaging and clinical data\u003c/h3\u003e\n\u003cp\u003eThree-phase enhancement images of the lesion, sex, age, CA199, CA125, CEA, BMI, abdominal pain, diabetes mellitus, hypertension, tumor site, metastasis, tumor size, morphology, boundary, texture, calcification, biliary dilatation, and lymph nodes, etc.\u003c/p\u003e\n\u003ch3\u003eMethod of examination\u003c/h3\u003e\n\u003cp\u003eA GE Discovery CT 750 multislice CT scanner was used to scan from the diaphragmatic apex to the level of the iliac spine. The scanning parameters were as follows: tube voltage of 120 kVp, tube current of 240\u0026thinsp;~\u0026thinsp;300 mA, layer thickness of 5 mm, reconstruction of 1.25 mm, layer spacing of 5 mm, and pitch of 1.375 : 1. The patient underwent CT scanning of the abdomen, followed by administration of the nonionic contrast agent iodohexol (300 mgI/ml) via the elbow vein mass at a dose of 1.5 ml/kg at a rate of 3.0 ml/s; the following steps were carried out in the sequence of the arterial (30 s), portal venous (70 s), and delayed (180\u0026thinsp;~\u0026thinsp;300 s) phases of the scan.\u003c/p\u003e\n\u003ch3\u003eLesion segmentation and feature extraction\u003c/h3\u003e\n\u003cp\u003ePreoperative enhanced CT images of the patient were collected and exported from the PACS workstation in DICOM format, and the axial continuous level images of the patient's three preoperative enhanced CT phases were selected for manual ROI outlining by two senior physicians who had been engaged in diagnostic CT imaging for more than 15 years. ITK-SNAP software (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) was used, and the two radiologists, both of whom were unaware of the histopathological results, avoided blood vessels during the outlining process. Calcification and other components were avoided during the outlining process, and when the outlining results were inconsistent, both radiologists made a joint decision via consultation. Two weeks later, the interclass and intraclass correlation efficiency (ICC) were calculated separately to assess observer agreement. Thirty patients were randomly selected from all patients with preoperative enhanced CT images, one of the doctors resketched the lesions to assess intraclass agreement, and the other doctor similarly sketched the same 30 lesions to assess interclass agreement. concordance, and the features with an ICC\u0026thinsp;\u0026ge;\u0026thinsp;0.75 were then analyzed in the next step. Feature extraction was performed via Image Standardization Initiative (IBSI)-compliant AK (3.3.0, GE Healthcare) software, and the types of features extracted included first-order statistical features, morphological features, grayscale covariance matrix features, grayscale tour features, and wavelet-variation and Laplace-variation features.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eUnsupervised consensus cluster analysis\u003c/h3\u003e\n\u003cp\u003eTo ensure data integrity and consistency, the clinical dataset was meticulously preprocessed.\u003c/p\u003e \u003cp\u003eMissing values, outliers and irregular data formats were corrected and adjusted. Categorical variables were processed via either solo thermal coding or ordered coding according to their characteristics, enabling them to be adapted to subsequent analytical models. Continuous variables, on the other hand, were standardized to ensure that they had a mean of 0 and a standard deviation of 1.\u003c/p\u003e \u003cp\u003eTo dig deeper into the internal structure and potential subgroups of the data, an unsupervised consensus clustering method was employed. Consensus clustering is an advanced technique for assessing the stability of clustering results. It provides important information about the stability of the clustering structure by clustering the dataset multiple times and observing the common belonging of samples in different clusters. The core element is that if two samples are frequently attributed to the same cluster over multiple clustering iterations, the consensus between them is greater. This relationship is explicitly described by generating a consensus matrix, where each element represents how often two samples are assigned to the same cluster across all clustering iterations. In addition, this matrix can be further clustered to obtain a robust cluster partitioning of the dataset. To find the optimal number of clusters (k), we use the covariance coefficient to evaluate the fitness of different k values. The covariance coefficient is a measure of the strength of the relationship between two variables that provides us with information about the stability of the cluster structure at each k value. Theoretically, a higher value of the coefficient of the symbiotic relationship indicates that the k value provides a better clustering structure for the dataset. By comparing the coefficients of symbiotic relationships at different values of k, we can determine an optimal value of k that maximizes that coefficient.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analyses and modeling\u003c/h2\u003e \u003cp\u003eAll the statistical analyses were performed in Python (version 3.8.2, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.python.org\u003c/span\u003e\u003cspan address=\"http://www.python.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). No statistical models were developed, and univariate logistic regression was performed for each clinical variable to assess its relationship with the clustered labels, fitting a logistic regression model with the clustered labels as the dependent variable and the clinical variables as the independent variables. Variables with P values less than 0.1 were selected for further analysis. Multivariate logistic regression models were fitted using the clinical variables and consensus clustering categories selected in the univariate analyses, Ridge regularization (L2 penalty) was used to prevent overfitting and address potential multicollinearity, with the strength of regularization (alpha) set to the default value of 1. The final logistic regression was constructed using the statistically significant clinical variables identified in the multivariate logistic regression analysis model, which was used to predict group labeling on the basis of individual clinical characteristics. The performance of the model was assessed via metrics such as the area under the receiver operating characteristic (ROC) curve. The coefficients of the final logistic regression model were interpreted for relationships between clinical variables, clusters and NF-pNETs/SPTs, and the predictive performance of the model is reported. The final p\u0026thinsp;\u0026lt;\u0026thinsp;0.2 clinical factors were then combined with the cluster model in a stepwise multivariate logistic regression analysis, leading to a joint model and presentation of the model in a column-line diagram. The predictive ability of the three models was evaluated and compared via the area under the curve (AUC) of the ROC curve as well as model accuracy, sensitivity, specificity and negative and positive predictive values (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStudy population\u003c/h2\u003e \u003cp\u003eFrom January 2016 to December 2022, 47 SPENTs and 45 SPTs were eligible for this retrospective study and were divided into a training set consisting of 17 males and 47 females (range 9\u0026ndash;76 years, mean 42.44 years) and a validation set consisting of 13 males and 15 females (range 11\u0026ndash;76 years, mean 46.89 years).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eClinical models\u003c/h2\u003e \u003cp\u003eA subset of clinical characteristics with strong covariates was removed. According to the results of the univariate analysis, the differences in hypertension, CEA, tumor boundary, tumor morphology, tumor texture and enhancement III were statistically significant (P\u0026thinsp;\u0026lt;\u0026thinsp;0.1), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. CEA, tumor boundary, tumor texture, arterial phase and delayed phase CT scans were shown to be clinically independent risk factors for disease development in the results of multifactorial logistic regression analysis (P\u0026thinsp;\u0026lt;\u0026thinsp;0.2), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of general information.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003eUnivariate Regression Analysis(P\u0026thinsp;\u0026lt;\u0026thinsp;0.1)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecoef\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003estd err\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ez\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eP\u0026gt;|z|\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e[0.025\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.975]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHypertension\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.401\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.716\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.956\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.804\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDiabetes\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9.338\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e51.561\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.856\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-91.720\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e110.396\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAbdominal Pain\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.366\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.506\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.724\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.469\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.626\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBMI\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-1.968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22.702\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-0.087\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.931\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-46.464\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e42.528\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCEA\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.933\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.334\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.790\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.588\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCA199\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.040\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.223\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.221\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.030\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.129\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCA125\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.015\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.713\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.476\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.019\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.040\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Size\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.046\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.296\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.068\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Boundary\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.092\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.640\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.838\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.346\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Morphology\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.601\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.049\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.002\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.654\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.008\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Texture\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.804\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.392\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.051\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.040\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.572\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-0.036\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBiliary Dilatation\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.597\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.684\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.873\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.383\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.939\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEalcification\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.680\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.549\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.239\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.215\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.395\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLymph Nodes\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.487\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e192.381\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.055\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.957\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-366.572\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e387.546\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConsensus_Group_A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.346\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.144\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.397\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.629\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-0.063\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConsensus_Group_D\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.127\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.080\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.472\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.027\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConsensus_Group_V\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.115\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.461\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.645\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.172\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.278\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMultifactor logistic regression analysis.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"7\" nameend=\"c7\" namest=\"c1\"\u003e \u003cp\u003eMultifactor regression analysis (P\u0026thinsp;\u0026lt;\u0026thinsp;0.2)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ecoef\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003estd err\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ez\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eP\u0026gt;|z|\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e[0.025\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.975]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConst\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.277\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.399\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.426\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.579\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHypertension\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.627\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.684\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.494\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.171\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.424\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCEA\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.672\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.365\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.840\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.044\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e1.387\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Boundary\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.971\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.822\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.068\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.134\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e3.671\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Morphology\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.636\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.871\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.730\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.465\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.071\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e2.344\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Texture\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.987\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.506\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.051\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-1.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.005\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConsensus_Group_A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.503\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.226\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-2.227\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.026\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.946\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-0.060\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eConsensus_Group_D\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.215\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-1.291\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.197\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e-0.542\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.111\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eUnsupervised consensus clustering model\u003c/h2\u003e \u003cp\u003eUnsupervised classification of lesions was performed via consensus clustering analysis incorporating the histological features of A, D, and V images, and the optimal numbers of clusters selected by the covariance coefficient, k, were 8, 4, and 7, respectively. The consensus clustering results were plotted according to the clustering results (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA). After one-way logistic regression analysis, the A and D consensus clustering categories had a significant relationship with outcome classification (P values\u0026thinsp;=\u0026thinsp;0.017 and 0.08, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Unsupervised cluster analysis for neuroendocrine tumors and solid pseudopapillary tumors was performed via a multifactorial logistic regression model incorporating A and D.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eJoint Clinical-Consensus Clustering Models\u003c/h2\u003e \u003cp\u003eThe most valuable clinical factors and unsupervised cluster analyses were used to establish a joint clinical cluster model via a multifactor logistic regression model. The predictive ability of the above three models was evaluated via model accuracy, sensitivity, specificity, and negative and positive predictive values (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), and the predictive ability of the above three models was compared via the area under the ROC curve (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). The AUC of the training centralized clinical model\u0026thinsp;=\u0026thinsp;0.81; the AUC increased to 0.88 after adding the cluster analysis results to build the joint model (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Delong test). Validation of the centralized clinical model yielded an AUC\u0026thinsp;=\u0026thinsp;0.66; after adding the results of cluster analysis to create a joint model, the AUC significantly increased to 0.85 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Delong test). This shows that the joint model yielded higher values than either the clinical model or the unsupervised clustering model did, which suggests that the joint model features play a highly significant role in discriminating between NF-pNETs and SPTs. The final generated model line diagrams and Noetherian diagrams are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eC and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA, B.\u003c/p\u003e \u003cp\u003eFinally, we assessed the clinical validity of the above models via DCA (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC), which revealed that the joint clinical-clustering model of unsupervised learning had a greater net benefit when the probability of a high-risk threshold was in the range of 0\u0026ndash;1, suggesting that the model had greater clinical validity than each of the other models.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance comparison of different models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC [95% CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eACC [95% CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSEN [95% CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSPE [95% CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePPV [95% CI]\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNPV [95% CI]\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003eClinical\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.81 [0.72\u0026ndash;0.90]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.781 [0.703\u0026ndash;0.859]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.788 [0.676\u0026ndash;0.900]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.774 [0.636\u0026ndash;0.893]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.788 [0.658\u0026ndash;0.900]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.774 [0.654\u0026ndash;0.893]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.66 [0.48\u0026ndash;0.83]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.679 [0.500\u0026ndash;0.821]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.857 [0.692\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.500 [0.273\u0026ndash;0.727]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.632 [0.429\u0026ndash;0.812]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.778 [0.500\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003eConsensus Groups\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.70 [0.59\u0026ndash;0.80]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.656 [0.562\u0026ndash;0.750]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.545 [0.400\u0026ndash;0.690]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.774 [0.647\u0026ndash;0.889]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.720 [0.560\u0026ndash;0.864]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.615 [0.486\u0026ndash;0.744]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.65 [0.47\u0026ndash;0.82]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.679 [0.536\u0026ndash;0.821]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.000 [1.000\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.357 [0.154\u0026ndash;0.583]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.609 [0.429\u0026ndash;0.773]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.000 [1.000\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003eNomogram\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88 [0.81\u0026ndash;0.94]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.812 [0.734\u0026ndash;0.891]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.848 [0.741\u0026ndash;0.946]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.774 [0.649\u0026ndash;0.886]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.800 [0.677\u0026ndash;0.903]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.828 [0.708\u0026ndash;0.941]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTesting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85 [0.71\u0026ndash;0.96]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.786 [0.643\u0026ndash;0.893]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.714 [0.500\u0026ndash;0.917]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.857 [0.692\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.833 [0.615\u0026ndash;1.000]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.750 [0.562\u0026ndash;0.929]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eSPTs and pNETs are both rare tumors of the pancreas, and the detection rate of both has gradually increased in recent years with the rapid development of medical imaging. The identification of NF-pNETs is more difficult than that of functional pNETs because of the absence of specific clinical symptoms. NF-pNETs are more aggressive because of their more invasive nature and more aggressive surgical approaches, including pancreatectomy and regional lymph node dissection, thereby achieving negative tumor margins[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Therefore, accurate preoperative diagnosis is important for the selection and guidance of clinical treatment. Currently, early screening of pancreatic tumors relies on imaging, and both NFPNETs and SPTs can present as well-defined round or round-like masses with solid, cystic, or cystic-solid borders, and enhancement can show insignificant or progressive enhancement[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. It is not always possible to accurately distinguish between the two on conventional imaging. Moreover, statistically significant differences in the clinical characteristics of patients with NF-pNETs and those with SPTs in terms of sex, age, calcification and tumor size have been reported in the literature, with NF-pNETs mostly observed in middle-aged and elderly people and SPTs being more common in young women[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]; however, it is more difficult to differentiate between atypical NF-PNETs and SPTs. Therefore, in this study, along with the basis of clinical factors combined with imaging histology, quantitative imaging histology cluster analysis will help to differentiate between the two and help clinicians plan surgery and choose combined treatment options.\u003c/p\u003e \u003cp\u003eIn this study, the clinical characteristics with strong covariates were first removed (sex and age) in the analyses targeting the clinical factors, indicating that these two factors differed significantly between SPTs and NF-PNETs. This difference occurred mainly in young women with a mean age of 34.3 years and in NF-PNETs, who were mainly in the middle-aged and older age groups with a mean age of 52.1 years, which is in line with the majority of the results in the literature. In addition, in the multifactorial analysis of clinical and imaging characteristics in this study, the CEA level, tumor border, tumor texture, presence of arterial-stage lesions and presence of delayed-stage lesions were found to be independent risk factors for differentiating NF-PNETs from SPTs. Most SPTs are relatively inert tumors with well-defined borders and intact perimeters, whereas NF-PNETs have no specific clinical manifestations, are found with larger lesions, are more likely to develop into malignant tumors, and exhibit invasive features such as blurred borders and incomplete perimeters, which is similar to the results of this study. The texture can affect the occurrence of both diseases. From the pathological point of view, solid pseudopapillary tumor tumor cells degenerate and shed to form cystic areas, the size of the cystic cavity varies, and the surrounding tumor cells are arranged in the form of trabecular or glandular vesicles. The tumor in the solid area is in the form of a sheet of lumps or diffusely distributed, and it is separated by small, fibrous blood vessels. The pseudopapillary area can be seen in the tumor cells surrounded by the surrounding fibrous blood vessels in the separation of the complex arrangement of forming a characteristic pseudo-pseudopapillary area, in which the tumor cells are surrounded by surrounding fibrous vessels to form a characteristic pseudopapillary structure[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]; NF-pNETs have a variety of histological structures under the microscope, and the cells can be arranged into solid nest-like, trabecular, gyrus-like, and glandular vesicles, etc., and may show cystic changes in the event of necrosis[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. The SPTs in this study were mainly cystic-solid and cystic, and the NF-PNETs were mainly solid, which is consistent with the pathological manifestations of these two types of tumors. The CT manifestations were based on the density differences presented by the pathological manifestations. In the present study, the enhancement of the arterial and delayed phases on enhanced CT was found to be meaningful in distinguishing SPTs from NF-PNETs. It is possible that NF-pNETs contain more abundant blood vessels and that the enhancement in the arterial phase is significant. SPT tumor cells are arranged in nested sheets and divided by small blood vessels into blood sinusoidal structures similar to those of cavernous hemangiomas, so the enhancement scans show progressive enhancement. In addition, CEA was statistically significant in this study, but its statistical significance was not mentioned in the relevant literature, which may be caused by the small sample size of this study.\u003c/p\u003e \u003cp\u003eCurrently, to circumvent the limitations of clinical features and imaging features, quantitative research methods such as imaging methods and imaging histology are carried out in the clinic. Most of the previous literature has studied the imaging features of SPTs and NF-PNETs, but this is too one-sided for atypical patients. In this study, on the one hand, we established an unsupervised clustering model, and the AUC values of the model in the training set and validation set were 0.70 (95% CI: 0.59\u0026ndash;0.80) and 0.65 (95% CI: 0.47\u0026ndash;0.82), respectively, with good predictive efficacy, which suggests that quantitative imaging of histological features can be used as a new way to identify SPTs and NFPNETs. On the other hand, when we combined the clinical factors and cluster analysis, the predictive efficacy was more significantly improved, with AUC values of 0.88 (95% CI: 0.81\u0026ndash;0.94) and 0.85 (95% CI: 0.71\u0026ndash;0.96) for the training and validation sets, respectively, suggesting that clinical factors are also particularly important for both tumors and that a combination of the two can maximize the benefits to patients. The combined clinical-clustering model in this study also provides a new method and idea for diagnosing SPTs and NF-PNETs.\u003c/p\u003e \u003cp\u003eIn addition, the clustering model developed in this study uses an unsupervised machine learning approach. Unsupervised consensus clustering is an advanced clustering technique for unsupervised learning. Unlike traditional one-shot clustering methods, this method clusters the data multiple times and in multiple ways and observes the common attributes of the samples across the clustering results. The main advantage of this method is its robustness to results, which significantly reduces biases introduced by random initializations or specific algorithms. In addition, it is robust and adaptable for handling a wide range of data types and structures, helping to account for subtle data structures that may have been overlooked in word clustering. Moreover, evaluating the consensus under different numbers of clusters can provide a strong guide for determining the optimal number of clusters.\u003c/p\u003e \u003cp\u003eThe limitations of this study are as follows: (i) The sample size is small, which may lead to model overfitting and have some impact on the stability of the model. (ii) Some disease-related laboratory indicators, such as neuron-specific enolase and chromogranin A, were not included. (iii) The evaluation and interpretation of unsupervised machine learning models may be relatively difficult due to the lack of labeling information. In addition, unsupervised learning algorithms usually require more computational resources and more complex algorithm designs to discover patterns efficiently from data.\u003c/p\u003e \u003cp\u003eIn summary, this study constructed an unsupervised clustering model based on enhanced CT combined with clinical features to construct a comprehensive model to distinguish between NF-pNETs and SPTs, which has great potential for identifying patients requiring monitoring or surgical resection.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eOur study found that unsupervised clustering models based on augmented CT potentially differentiate between SPTs and NF-PNETs, which can provide a new diagnostic basis for diagnostic imaging in the future.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAnnotation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eAK\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eArtificial Intelligence Kit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eArea Under the Curve\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eBody Mass Index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eCA125\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eCancer Antigen 125\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eCA199\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eCarbohydrate Antigen 19-9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eCEA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eCarcinoembryonic Antigen\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eConfidence Interval\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003ecoef\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eCoefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eComputed Tomography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eDICOM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eDigital Imaging and Communications in Medicine\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eGE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eGeneral Electric\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eICC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eIntraclass Correlation Coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eIBSI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eImage Biomarker Standardization Initiative\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eNF-pNETs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eNon-Functioning Pancreatic Neuroendocrine Tumors\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eNegative Predictive Value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003ePACS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003ePicture Archiving and Communication Systems\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003epNETs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003ePancreatic Neuroendocrine Tumors\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003ePositive Predictive Value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eSEN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eSPTs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eSolid Pseudopapillary Tumors\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003eSPE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 127px;\"\u003e\n \u003cp\u003estd err\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 332px;\"\u003e\n \u003cp\u003eStandard Error\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003e This study has been approved by the Institutional Ethics Review Committee of the First Affiliated Hospital of the University of Science and Technology of China (No. 2024-RE-464), and the Ethics Review Committee of the First Affiliated Hospital of the University of Science and Technology of China has waived informed individual consent for this retrospective analysis. All methods are executed in accordance with relevant regulations and guidelines.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eAuthors details\u003c/h2\u003e \u003cp\u003e \u003csup\u003e1\u003c/sup\u003eGraduate Department, Bengbu Medical University, Bengbu, Anhui, China\u003c/p\u003e \u003cp\u003e \u003csup\u003e2\u003c/sup\u003eDepartment of Radiology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Sciences and Technology of China, Hefei 230001, China\u003c/p\u003e \u003cp\u003e \u003csup\u003e3\u003c/sup\u003eGraduate Department, Wannan Medical College, Wuhu, Anhui, China\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis study was supported by grants from the National Natural Science Foundation of China (82271991) and the Anhui Province Key R\u0026amp;D Life and Health Category A Project (2022e0702007). No other potential conflicts of interest relevant to this article are reported.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eWZ and WW were responsible for research design and project management. PC was responsible for data collection. WC and MW were responsible for image processing and feature extraction. BW was responsible for article writing. DS and JL were responsible for statistical analysis. All authors contributed to the article and approved the submitted version.\u003c/p\u003e\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e \u003cp\u003eThe dataset generated for this study can be obtained from the correspondence author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWang Y, Miller FH, Chen ZE, Merrick L, Mortele KJ, Hoff FL, Hammond NA, Yaghmai V, Nikolaidis PJR: \u003cstrong\u003eDiffusion-weighted MR imaging of solid and cystic lesions of the pancreas\u003c/strong\u003e. 2011, \u003cstrong\u003e31\u003c/strong\u003e(3):E47-E64.\u003c/li\u003e\n\u003cli\u003eLow G, Panu A, Millo N, Leen EJR: \u003cstrong\u003eMultimodality imaging of neoplastic and nonneoplastic solid lesions of the pancreas\u003c/strong\u003e. 2011, \u003cstrong\u003e31\u003c/strong\u003e(4):993-1015.\u003c/li\u003e\n\u003cli\u003eNiederle B, Selberherr A, Bartsch DK, Brandi ML, Doherty GM, Falconi M, Goudet P, Halfdanarson TR, Ito T, Jensen RTJN: \u003cstrong\u003eMultiple endocrine neoplasia type 1 and the pancreas: diagnosis and treatment of functioning and non-functioning pancreatic and duodenal neuroendocrine neoplasia within the MEN1 syndrome\u0026ndash;an international consensus statement\u003c/strong\u003e. 2021, \u003cstrong\u003e111\u003c/strong\u003e(7):609-630.\u003c/li\u003e\n\u003cli\u003eWu J, Sun C, Li E, Wang J, He X, Yuan R, Yi C, Liao W, Wu LJBc: \u003cstrong\u003eNon-functional pancreatic neuroendocrine tumours: emerging trends in incidence and mortality\u003c/strong\u003e. 2019, \u003cstrong\u003e19\u003c/strong\u003e:1-10.\u003c/li\u003e\n\u003cli\u003eBrook OR, Gourtsoyianni S, Brook A, Siewert B, Kent T, Raptopoulos VJR: \u003cstrong\u003eSplit-bolus spectral multidetector CT of the pancreas: assessment of radiation dose and tumor conspicuity\u003c/strong\u003e. 2013, \u003cstrong\u003e269\u003c/strong\u003e(1):139-148.\u003c/li\u003e\n\u003cli\u003eMarin D, Nelson RC, Barnhart H, Schindera ST, Ho LM, Jaffe TA, Yoshizumi TT, Youngblood R, Samei EJR: \u003cstrong\u003eDetection of pancreatic tumors, image quality, and radiation dose during the pancreatic parenchymal phase: effect of a low-tube-voltage, high-tube-current CT technique\u0026mdash;preliminary results\u003c/strong\u003e. 2010, \u003cstrong\u003e256\u003c/strong\u003e(2):450-459.\u003c/li\u003e\n\u003cli\u003eScialpi M, Reginelli A, D\u0026apos;Andrea A, Gravante S, Falcone G, Baccari P, Manganaro L, Palumbo B, Cappabianca SJIjos: \u003cstrong\u003ePancreatic tumors imaging: An update\u003c/strong\u003e. 2016, \u003cstrong\u003e28\u003c/strong\u003e:S142-S155.\u003c/li\u003e\n\u003cli\u003eLambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, Van Stiphout RG, Granton P, Zegers CM, Gillies R, Boellard R, Dekker AJEjoc: \u003cstrong\u003eRadiomics: extracting more information from medical images using advanced feature analysis\u003c/strong\u003e. 2012, \u003cstrong\u003e48\u003c/strong\u003e(4):441-446.\u003c/li\u003e\n\u003cli\u003eMukherjee S, Patra A, Khasawneh H, Korfiatis P, Rajamohan N, Suman G, Majumder S, Panda A, Johnson MP, Larson NBJG: \u003cstrong\u003eRadiomics-based machine-learning models can detect pancreatic cancer on prediagnostic computed tomography scans at a substantial lead time before clinical diagnosis\u003c/strong\u003e. 2022, \u003cstrong\u003e163\u003c/strong\u003e(5):1435-1446. e1433.\u003c/li\u003e\n\u003cli\u003eBr\u0026ouml;ker F, Holt LL, Roads BD, Dayan P, Love BCJTiCS: \u003cstrong\u003eDemystifying unsupervised learning: how it helps and hurts\u003c/strong\u003e. 2024.\u003c/li\u003e\n\u003cli\u003eTobaly D, Santinha J, Sartoris R, Dioguardi Burgio M, Matos C, Cros J, Couvelard A, Rebours V, Sauvanet A, Ronot MJC: \u003cstrong\u003eCT-based radiomics analysis to predict malignancy in patients with intraductal papillary mucinous neoplasm (IPMN) of the pancreas\u003c/strong\u003e. 2020, \u003cstrong\u003e12\u003c/strong\u003e(11):3089.\u003c/li\u003e\n\u003cli\u003eWan Y, Yang P, Xu L, Yang J, Luo C, Wang J, Chen F, Wu Y, Lu Y, Ruan DJMP: \u003cstrong\u003eRadiomics analysis combining unsupervised learning and handcrafted features: A multiple‐disease study\u003c/strong\u003e. 2021, \u003cstrong\u003e48\u003c/strong\u003e(11):7003-7015.\u003c/li\u003e\n\u003cli\u003eLock EF, Dunson DBJB: \u003cstrong\u003eBayesian consensus clustering\u003c/strong\u003e. 2013, \u003cstrong\u003e29\u003c/strong\u003e(20):2610-2616.\u003c/li\u003e\n\u003cli\u003eAkirov A, Larouche V, Alshehri S, Asa SL, Ezzat SJC: \u003cstrong\u003eTreatment options for pancreatic neuroendocrine tumors\u003c/strong\u003e. 2019, \u003cstrong\u003e11\u003c/strong\u003e(6):828.\u003c/li\u003e\n\u003cli\u003eKatsoulakis E, Yu Y, Apte AP, Leeman JE, Katabi N, Morris L, Deasy JO, Chan TA, Lee NY, Riaz NJOo: \u003cstrong\u003eRadiomic analysis identifies tumor subtypes associated with distinct molecular and microenvironmental factors in head and neck squamous cell carcinoma\u003c/strong\u003e. 2020, \u003cstrong\u003e110\u003c/strong\u003e:104877.\u003c/li\u003e\n\u003cli\u003eWang C, Cui W, Wang J, Chen X, Tong H, Wang ZJAR: \u003cstrong\u003eDifferentiation between solid pseudopapillary neoplasm of the pancreas and hypovascular pancreatic neuroendocrine tumors by using computed tomography\u003c/strong\u003e. 2019, \u003cstrong\u003e60\u003c/strong\u003e(10):1216-1223.\u003c/li\u003e\n\u003cli\u003eLiu F, Li J, Fang X, Meng Y, Zhang H, Yu J, Feng X, Wang L, Jiang H, Lu JJAR: \u003cstrong\u003eDifferentiation of Solid Pseudopapillary Tumor and Non-Functional Neuroendocrine Tumors of the Pancreas Based on CT Delayed Imaging: A Propensity Score Analysis\u003c/strong\u003e. 2022, \u003cstrong\u003e29\u003c/strong\u003e(3):350-357.\u003c/li\u003e\n\u003cli\u003eHao EIU, Hwang HK, Yoon D-S, Lee WJ, Kang CMJM: \u003cstrong\u003eAggressiveness of solid pseudopapillary neoplasm of the pancreas: a literature review and meta-analysis\u003c/strong\u003e. 2018, \u003cstrong\u003e97\u003c/strong\u003e(49):e13147.\u003c/li\u003e\n\u003cli\u003ePathology DoDDJCJo: \u003cstrong\u003eChinese consensus on the pathological diagnosis of gastrointestinal and pancreatic neuroendocrine neoplasms (2020)(in Chinese)\u003c/strong\u003e. 2021, \u003cstrong\u003e50\u003c/strong\u003e:14-20.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Imaging histology, Unsupervised learning, Consensus clustering, Pancreatic nonfunctional neuroendocrine tumor, Pancreatic solid pseudopapillary tumor","lastPublishedDoi":"10.21203/rs.3.rs-5782491/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5782491/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e: Unsupervised clustering has played a greater role in the diagnosis and differential diagnosis of pancreatic tumors in recent years. This study aimed to investigate the value of constructing a c\u003ca href=\"#_ftn1\" title=\"\"\u003e[1]\u003c/a\u003elustering model for unsupervised learning based on enhanced CT to identify pancreatic nonfunctional neuroendocrine tumors (NF-pNETs) and solid pseudopapillary tumors (SPTs).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: 45 patients with SPTs and 47 patients with NF-psNETs were retrospectively analyzed. The data were randomly divided into a training set and a validation set at a ratio of 7:3. One-way logistic regression was performed for each clinical variable to assess its relationship with the clustered labels, and a logistic regression model was fitted with the clustered labels as the dependent variable and the clinical variables as independent variables. Variables with P values \u0026lt;0.1 were selected for further analysis. Multifactorial logistic regression models were fitted via the clinical variables selected in the univariate analysis, ridge regularization (L2 penalty) was used to prevent overfitting and address potential multicollinearity, with the strength of regularization (alpha) set to a default value of 1. Clinical variables that were meaningful in the multifactorial logistic regression analyses were used to construct the final logistic regression model, which was used to predict individual clinical-based group labeling on the basis of individual clinical characteristics. For imaging, the optimal number of clusters k selected by the covariance coefficient was used to build the clustering model via unsupervised classification of the lesions through consensus clustering analysis fusing the imaging histological features of arterial-phase, venous-phase, and delayed-phase images. The clinical factors with a final p value \u0026lt; 0.2 were subsequently combined with the clustering model to perform stepwise multivariate logistic regression analyses, thereby establishing a joint model. The performance of the three models was assessed via AUC values, and column line plots were generated to visualize the models. Finally, the clinical validity of the models was assessed via decision curve analysis (DCA).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e:\u0026nbsp; The AUCs of the clustered and clinical models were 0.70 and 0.74 (95% CI: 0.66–0.81) in the training set and 0.65 and 0.75 (95% CI: 0.64–0.87) in the validation set, and the AUCs of the training and test sets in the joint model were 0.88 (95% CI: 0.81–0.94) and 0.85 (95% CI: 0.71–0.96), respectively. Decision curve analysis revealed that the joint clinical-clustering model had a greater net benefit when the high-risk threshold probability was in the range of 0--1.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions\u003c/strong\u003e: Unsupervised clustering models based on enhanced CT have potential for discriminating between SPTs and NF-PNETs, which can inform clinical decision-making.\u003c/p\u003e","manuscriptTitle":"Identification of pancreatic nonfunctional neuroendocrine tumors and solid pseudopapillary tumors via the construction of a consensus clustering model based on enhanced CT images","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-13 12:57:38","doi":"10.21203/rs.3.rs-5782491/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b8f35ae4-039f-44b5-b21f-d63694f66cb5","owner":[],"postedDate":"January 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-01-13T13:05:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-13 12:57:38","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5782491","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5782491","identity":"rs-5782491","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}