Modela
|
RFC (Model 1)b
|
SVM (Model 2)b
|
RFC (Model 3)c
|
SVM (Model 4)c
|
---|
Number of featuresd
|
2
|
18
|
28
|
66
|
Train AUC
|
0.95 (0.91–0.98)
|
0.88 (0.81–0.94)
|
0.84 (0.82–0.85)
|
0.91 (0.88–0.94)
|
Test AUC
|
0.90 (0.85–0.95)
|
0.76 (0.67–0.84)
|
0.88 (0.84–0.91)
|
0.90 (0.87–0.93)
|
Sensitivitye
|
86%
|
93%
|
86%
|
83%
|
Specificitye
|
78%
|
31%
|
92%
|
95%
|
NPVe
|
89%
|
73%
|
81%
|
78%
|
PPV e
|
73%
|
69%
|
94%
|
96%
|
- a Optimized for the classification of low-grade vs. high-grade fibrosis in liver segments;
- b The liver segments were randomly divided into equal size training and test set;
- c Segments of patients who had been scanned with a 64-slice scanner constituted the training set, and patients who had been scanned with a 16-slice scanner were assigned to the test set;
- dAfter cross-validated recursive feature elimination;
- e Calculated in the test set; RFC random forest classifier, SVM support vector machine classifier, NPV negative predictive value, PPV positive predictive value