Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 May 23:14:1359364.
doi: 10.3389/fonc.2024.1359364. eCollection 2024.

Construction and validation of an endoscopic ultrasonography-based ultrasomics nomogram for differentiating pancreatic neuroendocrine tumors from pancreatic cancer

Affiliations

Construction and validation of an endoscopic ultrasonography-based ultrasomics nomogram for differentiating pancreatic neuroendocrine tumors from pancreatic cancer

Shuangyang Mo et al. Front Oncol. .

Abstract

Objectives: To develop and validate various ultrasomics models based on endoscopic ultrasonography (EUS) for retrospective differentiating pancreatic neuroendocrine tumors (PNET) from pancreatic cancer.

Methods: A total of 231 patients, comprising 127 with pancreatic cancer and 104 with PNET, were retrospectively enrolled. These patients were randomly divided into either a training or test cohort at a ratio of 7:3. Ultrasomics features were extracted from conventional EUS images, focusing on delineating the region of interest (ROI) for pancreatic lesions. Subsequently, dimensionality reduction of the ultrasomics features was performed by applying the Mann-Whitney test and least absolute shrinkage and selection operator (LASSO) algorithm. Eight machine learning algorithms, namely logistic regression (LR), light gradient boosting machine (LightGBM), multilayer perceptron (MLP), random forest (RF), extra trees, k nearest neighbors (KNN), support vector machine (SVM), and extreme gradient boosting (XGBoost), were employed to train prediction models using nonzero coefficient features. The optimal ultrasomics model was determined using a ROC curve and utilized for subsequent analysis. Clinical-ultrasonic features were assessed using both univariate and multivariate logistic regression. An ultrasomics nomogram model, integrating both ultrasomics and clinical-ultrasonic features, was developed.

Results: A total of 107 EUS-based ultrasomics features were extracted, and 6 features with nonzero coefficients were ultimately retained. Among the eight ultrasomics models based on machine learning algorithms, the RF model exhibited superior performance with an AUC= 0.999 (95% CI 0.9977 - 1.0000) in the training cohort and an AUC= 0.649 (95% CI 0.5215 - 0.7760) in the test cohort. A clinical-ultrasonic model was established and evaluated, yielding an AUC of 0.999 (95% CI 0.9961 - 1.0000) in the training cohort and 0.847 (95% CI 0.7543 - 0.9391) in the test cohort. Subsequently, the ultrasomics nomogram demonstrated a significant improvement in prediction accuracy in the test cohort, as evidenced by an AUC of 0.884 (95% CI 0.8047 - 0.9635) and confirmed by the Delong test. The calibration curve and decision curve analysis (DCA) depicted this ultrasomics nomogram demonstrated superior accuracy. They also yielded the highest net benefit for clinical decision-making compared to alternative models.

Conclusions: A novel ultrasomics nomogram was proposed and validated, that integrated clinical-ultrasonic and ultrasomics features obtained through EUS, aiming to accurately and efficiently identify pancreatic cancer and PNET.

Keywords: endoscopic ultrasonography; machine learning; nomogram; pancreatic cancer; pancreatic neuroendocrine tumors; ultrasomics.

PubMed Disclaimer

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Figures

Figure 1
Figure 1
Flowchart for enrolling the study population.
Figure 2
Figure 2
The workflow for this study.
Figure 3
Figure 3
The forest map of univariate logistic regression of clinical and ultrasonic features. *** means P < 0.001.
Figure 4
Figure 4
Ultrasomics feature selection with the LASSO regression model. (A) The LASSO model’s tuning parameter (λ) selection used 10-fold cross-validation via minimum criterion. The vertical lines illustrate the optimal value of the LASSO tuning parameter (λ). (B) LASSO coefficient profile plot with different log (λ) was displayed. The vertical dashed lines represent 6 ultrasomics features with nonzero coefficients selected with the optimal λ value.
Figure 5
Figure 5
The bar graph of 6 ultrasomics features that achieved nonzero coefficients.
Figure 6
Figure 6
The ROC curves of different ultrasomics models based on eight machine-learning algorithms for predicting PNET. (A) The ROC curves of different ultrasomics models in the training cohort. (B) The ROC curves of different ultrasomics models in the test cohort.
Figure 7
Figure 7
The ROC curves of the ultrasomics nomogram model (abbreviated “Nomogram”), clinical-ultrasonic signature (abbreviated “Clinic Signature”), and ultrasomics signature (abbreviated “Ultra Signature”) in the (A) training cohort and (B) test cohort, respectively.
Figure 8
Figure 8
The ultrasomics nomogram model predicts PNET based on clinical-ultrasonic signature (abbreviated “Clinic_Sig”) and ultrasomics signatures (abbreviated “Ultra_Sig”) simultaneously. The nomogram is used by summing all points identified on the scale for each variable. The total points projected on the bottom scales indicate the probabilities of PNET.
Figure 9
Figure 9
The calibration curves for the ultrasomics nomogram model (abbreviated “Nomogram”), clinical-ultrasonic signature (abbreviated “Clinic Signature”), and ultrasomics signature (abbreviated “Ultra Signature”) in the (A) training cohort and (B) test cohort, respectively.
Figure 10
Figure 10
The DCA curves for the ultrasomics nomogram model (abbreviated “Nomogram”), clinical-ultrasonic signature (abbreviated “Clinic Signature”), and ultrasomics signature (abbreviated “Ultra Signature”) in the (A) training cohort and (B) test cohort, respectively.

Similar articles

Cited by

References

    1. Scott AT, Weitz M, Breheny PJ, Ear PH, Darbro B, Brown BJ, et al. . Gene expression signatures identify novel therapeutics for metastatic pancreatic neuroendocrine tumors. Clin Cancer Res. (2020) 26:2011–21. doi: 10.1158/1078-0432.CCR-19-2884 - DOI - PMC - PubMed
    1. Song Y, Gao S, Tan W, Qiu Z, Zhou H, Zhao Y. Multiple machine learnings revealed similar predictive accuracy for prognosis of PNETs from the surveillance, epidemiology, and end result database. J Cancer. (2018) 9:3971–8. doi: 10.7150/jca.26649 - DOI - PMC - PubMed
    1. Keutgen XM, Kumar S, Gara SK, Boufraqech M, Agarwal S, Hruban RH, et al. . Transcriptional alterations in hereditary and sporadic nonfunctioning pancreatic neuroendocrine tumors according to genotype. Cancer. (2018) 124:636–47. doi: 10.1002/cncr.31057 - DOI - PMC - PubMed
    1. Li D, Zhou Y, Liu Y, Lin Y, Yu M, Lu X, et al. . Decreased expression of LKB1 predicts poor prognosis in pancreatic neuroendocrine tumor patients undergoing curative resection. OncoTargets Ther. (2018) 11:1259–65. doi: 10.2147/OTT - DOI - PMC - PubMed
    1. Frost M, Lines KE, Thakker RV. Current and emerging therapies for PNETs in patients with or without MEN1. Nat Rev Endocrinol. (2018) 14:216–27. doi: 10.1038/nrendo.2018.3 - DOI - PMC - PubMed

LinkOut - more resources