Table 1 CNN2 performance assessment.

From: Machine learning outperforms clinical experts in classification of hip fractures

 

Actual

Total

No fracture

Trochanteric

Intracapsular

Predicted

No fracture

304

12

13

329

Trochanteric

1

169

6

176

Intracapsular

15

14

198

227

Total

320

195

217

 

Precision

0.92

0.96

0.87

 

95% CI

0.89 to 0.95

0.92 to 0.98

0.82 to 0.91

 

Recall

0.95

0.87

0.91

 

95% CI

0.92 to 0.97

0.81 to 0.91

0.87 to 0.95

 

F1

0.94

0.91

0.89

 
  1. Precision = (number correctly predicted as class A)/(number predicted as class A). Recall = (number correctly predicted as class A)/(number actually of class A). F1 varies from 1 = perfect classifier for class A, to 0 = no image was correctly identified as class A.
Back to article page