Table 4 Diagnostic performance of the TWIST algorithm
From: A machine learning approach for differentiating malignant from benign enhancing foci on breast MRI
Training/testing sets | ||||
|---|---|---|---|---|
A/B | B/A | Total | 95% confidence interval (%) | |
Sensitivity | 100% | 100% | 100% | 87–100 |
Specificity | 90% | 91% | 90% | 77–97 |
Accuracy | 94% | 95% | 94% | 86–98 |
Positive predictive value | 85% | 89% | 87% | 70–96 |
Negative predictive value | 100% | 100% | 100% | 91–100 |
True positives | 11 | 16 | 27 | |
True negatives | 18 | 19 | 37 | |
False positives | 2 | 2 | 4 | |
False negatives | 0 | 0 | 0 | |
Positive likelihood ratio | 10 | 11 | 10 | |
Negative likelihood ratio | 0 | 0 | 0 | |
Area under the curve | 0.93 | 0.95 | ||
0.94* | ||||