Table 2 Diagnostic performance matrix.

From: Detection of acute thoracic aortic dissection based on plain chest radiography and a residual neural network (Resnet)

 

Internal validation

 

Onefold

Actual positive

Actual negative

Twofold

Actual positive

Actual negative

0.5 of Cut-off Value

Predicted positive

113

23

Predicted positive

112

31

Predicted negative

20

(Type A 19/117, 16.3%; Type B 1/16, 6.2%)

470

Predicted negative

21 (Type A 17/114, 14.9%; Type B 4/19, 11.1%)

462

 

Internal validation

 

Threefold

Actual positive

Actual negative

Fourfold

Actual positive

Actual negative

0.5 of Cut-off Value

Predicted positive

108

34

Predicted positive

119

38

Predicted negative

24(Type A 21/105, 20.0%; Type B 3/27, 11.1%)

459

Predicted negative

13 (Type A 10/111, 9.1%; Type B 3/21, 14.3%)

455

 

Internal validation

Test

Fivefold

Actual positive

Actual negative

 

Actual positive

Actual negative

0.5 of Cut-off Value

Predicted positive

117

38

Predicted positive

51

17

Predicted negative

15 (Type A 14/109, 12.8%; Type B 1/23, 4.3%)

455

Predicted negative

3 (Type A 0/31, 0.0%; Type B 3/23, 13.1%)

133

  1. Fivefold validation was performed with the dataset from hospital A and B. A hard voting method was used with five models from fivefold cross-validation to obtain the final classification result. Type A, type A aortic dissection; Type B, type B aortic dissection.
Back to article page