Predicting Mismatch-Repair Status in Rectal Cancer Using Multiparametric MRI-Based Radiomics Models: A Preliminary Study
Table 3
ROC analysis of the radiomics models.
ModelT2WI
ModelDWI
ModelCE-T1WI
Modelcombination
Training set
AUC
0.768
0.670
0.718
0.910
95% CI
0.659-0.856
0.554-0.772
0.605-0.814
0.823-0.963
Sensitivity
0.844
0.328
0.844
0.844
Specificity
0.643
1.000
0.643
0.929
Accuracy
0.808
0.449
0.808
0.859
NRI
0.285
0.444
0.285
/
value
0.028
0.043
0.002
/
Test set
AUC
0.707
0.568
0.691
0.901
95% CI
0.523-0.852
0.385-0.738
0.507-0.840
0.746-0.977
Sensitivity
0.407
0.667
0.926
1.000
Specificity
1.000
0.667
0.500
0.667
Accuracy
0.515
0.667
0.848
0.939
NRI
0.260
0.334
0.241
/
value
0.019
0.042
0.042
/
Validation set
AUC
0.721
0.603
0.702
0.874
95% CI
0.595-0.825
0.474-0.722
0.576-0.809
0.768-0.943
Sensitivity
0.909
0.364
0.909
0.909
Specificity
0.667
0.926
0.704
0.815
Accuracy
0.708
0.831
0.738
0.831
NRI
0.148
0.434
0.111
/
value
0.004
0.025
0.001
/
Modelcombination: based on the combination of multisequences. NRI: net reclassification improvement, Modelcombination compared with other models. Compared with Modelcombination by DeLong test.
