MR-Based Radiomics for Differential Diagnosis between Cystic Pituitary Adenoma and Rathke Cleft Cyst

Table 3

Comparison of diagnostic performance of the semantic model, multiparametric model, and combined radiomics and semantic model using ANN classifier in the training and test set.

Training set	AUC	Sensitivity	Specificity	Accuracy	PPV	NPV
Semantic model	0.902 [86.3, 94.1]	0.756 [66.3, 84.9]	0.933 [88.2, 98.5]	0.849 [79.5, 90.2]	0.912 [84.4, 97.9]	0.808 [73.2, 88.3]
Multiparametric model	0.89 [85.1, 92.9]	0.793 [70.5, 88.0]	0.844 [77.0, 91.9]	0.820 [76.2, 87.7]	0.823 [73.9, 90.7]	0.817 [73.9, 89.6]
Combined radiomics and semantic model	0.924 [88.5, 96.3]	0.866 [79.2, 94.0]	0.844 [77.0, 91.9]	0.855 [80.2, 90.7]	0.835 [75.6, 91.4]	0.874 [80.4, 94.3]
Test set	AUC	Sensitivity	Specificity	Accuracy	PPV	NPV
Semantic model	0.823 [70.5, 94.1]	0.783 [61.4, 95.1]	0.850 [69.4, 100.0]	0.814 [69.8, 93.0]	0.857 [70.7, 100.0]	0.773 [59.8, 94.8]
Multiparametric model	0.792 [67.4, 91.0]	0.522 [31.8, 72.6]	0.900 [76.9, 100.0]	0.698 [56.0, 83.5]	0.857 [67.4, 100.0]	0.621 [44.4, 79.7]
Combined radiomics and semantic model	0.848 [75.0, 94.6]	0.739 [56.0, 91.9]	0.800 [62.5, 97.5]	0.767 [64.1, 89.4]	0.810 [64.2, 97.7]	0.727 [54.1, 91.3]

Note: data in parentheses are 95% confidence intervals. PPV: positive predict value; NPV: negative predict value.