Quantitative analysis of enhanced CT in differentiating well-differentiated pancreatic neuroendocrine tumors and poorly differentiated neuroendocrine carcinomas

Machine Learning Methods Based on CT Features Differentiate G1/G2 From G3 Pancreatic Neuroendocrine Tumors

RATIONALE AND OBJECTIVES

To identify CT features for distinguishing grade 1 (G1)/grade 2 (G2) from grade 3 (G3) pancreatic neuroendocrine tumors (PNETs) using different machine learning (ML) methods.

MATERIALS AND METHODS

A total of 147 patients with 155 lesions confirmed by pathology were retrospectively included. Clinical-demographic and radiological CT features was collected. The entire cohort was separated into training and validation groups at a 7:3 ratio. Least absolute shrinkage and selection operator (LASSO) algorithm and principal component analysis (PCA) were used to select features. Three ML methods, namely logistic regression (LR), support vector machine (SVM), and K-nearest neighbor (KNN) were used to build a differential model. Receiver operating characteristic (ROC) curves and precision-recall curves for each ML method were generated. The area under the curve (AUC), accuracy rate, sensitivity, and specificity were calculated.

RESULTS

G3 PNETs were more likely to present with invasive behaviors and lower enhancement than G1/G2 PNETs. The LR classifier yielded the highest AUC of 0.964 (95% confidence interval [CI]: 0.930, 0.972), with 95.4% accuracy rate, 95.7% sensitivity, and 92.9% specificity, followed by SVM (AUC: 0.957) and KNN (AUC: 0.893) in the training group. In the validation group, the SVM classier reached the highest AUC of 0.952 (95% CI: 0.860, 0.981), with 91.5% accuracy rate, 97.3% sensitivity, and 70% specificity, followed by LR (AUC: 0.949) and KNN (AUC: 0.923).

CONCLUSIONS

The LR and SVM classifiers had the best performance in the training group and validation group, respectively. ML method could be helpful in differentiating between G1/G2 and G3 PNETs.

Clinical and CT Quantitative Features for Predicting Liver Metastases in Patients with Pancreatic Neuroendocrine Tumors: A Study with Prospective/External Validation

RATIONALE AND OBJECTIVES

We aimed to evaluate clinical characteristics and quantitative CT imaging features for the prediction of liver metastases (LMs) in patients with pancreatic neuroendocrine tumors (PNETs).

METHODS

Patients diagnosed with pathologically confirmed PNETs were included, 133 patients were in the training group, 22 patients in the prospective internal validation group, and 28 patients in the external validation group. Clinical information and quantitative features were collected. The independent variables for predicting LMs were confirmed through the implementation of univariate and multivariate logistic analyses. The diagnostic performance was evaluated by conducting receiver operating characteristic curves for predicting LMs in the training and validation groups.

RESULTS

PNETs with LMs demonstrated significantly larger diameter and lower arterial/portal tumor-parenchymal enhancement ratio, arterial/portal absolute enhancement value (AAE/PAE value) (p < 0.05). After multivariate analyses, A high level of tumor marker (odds ratio (OR): 5.32; 95% CI, 1.54-18.35), maximum diameter larger than 24.6 mm (OR: 7.46; 95% CI, 1.70-32.72), and AAE value ≤ 51 HU (OR: 4.99; 95% CI, 0.93-26.95) were independent positive predictors of LMs in patients with PNETs, with area under curve (AUC) of 0.852 (95%CI, 0.781-0.907). The AUCs for prospective internal and external validation groups were 0.883 (95% CI, 0.686-0.977) and 0.789 (95% CI, 0.602-0.916), respectively.

CONCLUSION

Tumor marker, maximum diameter and absolute enhancement value in arterial phase were independent predictors with good predictive performance for the prediction of LMs in patients with PNETs. Combining clinical and quantitative features may facilitate the attainment of good predictive precision in predicting LMs.

Duodenal neuroendocrine neoplasms on enhanced CT: establishing a diagnostic model with duodenal gastrointestinal stromal tumors in the non-ampullary area and analyzing the value of predicting prognosis

OBJECTIVE

To identify CT features and establish a diagnostic model for distinguishing non-ampullary duodenal neuroendocrine neoplasms (dNENs) from non-ampullary duodenal gastrointestinal stromal tumors (dGISTs) and to analyze overall survival outcomes of all dNENs patients.

MATERIALS AND METHODS

This retrospective study included 98 patients with pathologically confirmed dNENs (n = 44) and dGISTs (n = 54). Clinical data and CT characteristics were collected. Univariate analyses and binary logistic regression analyses were performed to identify independent factors and establish a diagnostic model between non-ampullary dNENs (n = 22) and dGISTs (n = 54). The ROC curve was created to determine diagnostic ability. Cox proportional hazards models were created and Kaplan-Meier survival analyses were performed for survival analysis of dNENs (n = 44).

RESULTS

Three CT features were identified as independent predictors of non-ampullary dNENs, including intraluminal growth pattern (OR 0.450; 95% CI 0.206-0.983), absence of intratumoral vessels (OR 0.207; 95% CI 0.053-0.807) and unenhanced lesion > 40.76 HU (OR 5.720; 95% CI 1.575-20.774). The AUC was 0.866 (95% CI 0.765-0.968), with a sensitivity of 90.91% (95% CI 70.8-98.9%), specificity of 77.78% (95% CI 64.4-88.0%), and total accuracy rate of 81.58%. Lymph node metastases (HR: 21.60), obstructive biliary and/or pancreatic duct dilation (HR: 5.82) and portal lesion enhancement ≤ 99.79 HU (HR: 3.02) were independent prognostic factors related to poor outcomes.

CONCLUSION

We established a diagnostic model to differentiate non-ampullary dNENs from dGISTs. Besides, we found that imaging features on enhanced CT can predict OS of patients with dNENs.

The maximal contrast-enhanced range of CT for differentiating the WHO pathological subtypes and risk subgroups of thymic epithelial tumors

OBJECTIVE

To explore the value of maximal contrast-enhanced (CEmax) range using contrast-enhanced CT (CECT) imaging in differentiating the pathological subtypes and risk subgroups of thymic epithelial tumors (TETs).

METHODS

The pre-treatment-CECT images of 319 TET patients from May 2012 to November 2021 were analyzed retrospectively. The CEmax was defined as the maximum difference between the CT value of the solid tumor on pre-contrast and contrast-enhanced images. The mean CEmax value was calculated at three different tumor levels.

RESULTS

There was a significant difference in the CEmax among the eight main pathological subtypes [types A, AB, B1, B2, and B3 thymoma, thymic carcinoma (TC), low-grade neuroendocrine tumor (NET) and high-grade NET] (p < 0.001). Among the eight subtypes, the CEmax values of types A, AB, and low-risk NET were higher than those of the other subtypes (all p < 0.001), and there was no difference among types B1-B3 and high-risk NET (all p > 0.05). There was no difference for CEmax values between NET and TC (p = 0.491). For the risk subgroups, the CEmax of TC (including NET) was 35.35 ± 11.41 HU, which was lower than that of low-risk thymoma (A and AB) (57.73±21.24 HU) (P < 0.001) and was higher than that of high-risk thymoma (B1-B3) (27.37±8.27 HU) (P < 0.001). The CEmax cut-off values were 38.5 HU and 30.5 HU respectively (AUC: 0.829 and 0.712; accuracy, 72.4% and 67.7%).

CONCLUSION

The tumor CEmax on CECT helps differentiate the pathological subtypes and risk subgroups of TETs.

ADVANCES IN KNOWLEDGE

In this study, an improved simplified risk grouping method was proposed based on the traditional (2004 edition) simplified risk grouping method for TETs. If Type B1 thymoma is classified as high-risk, radiologists using this improved method may improve the accuracy in differentiating risk level of TETs compared with the traditional method.