Table 3 Deep learning models outperform logistic regression (LR) under the constraint that sensitivity is no less than 90%.

From: Identification of sentinel lymph node macrometastasis in breast cancer by deep learning based on clinicopathological characteristics

 

T-size

Multivariable models

LR

MLP

Resnet

Transformer

CatBoost

ROC AUC

0.692

0.711 (± 0.002)

0.712 (± 0.002)

0.708 (± 0.003)

0.711 (± 0.004)

0.704 (± 0.004)

PR AUC

0.239

0.273 (± 0.001)

0.263 (± 0.004)

0.253 (± 0.012)

0.267 (± 0.010)

0.258 (± 0.008)

Sensitivity (recall TPR), %

90.1

90.1

90.1

90.1

90.1

90.1

Specificity (TNR), %

31.8

32.6 (± 0.5)

34.2 (± 0.9)

33.0 (± 0.7)

34.6 (± 0.6)

32.8 (± 1.3)

PPV (precision), %

15.6

15.8 (± 0.1)

16.1 (± 0.2)

15.9 (± 0.1)

16.2 (± 0.1)

15.8 (± 0.2)

NPV, %

95.8

95.9 (± 0.1)

96.1 (± 0.1)

96.0 (± 0.1)

96.2 (± 0.1)

95.9 (± 0.2)

Accuracy, %

39.0

39.7 (± 0.4)

41.0 (± 0.8)

40.0 (± 0.6)

41.5 (± 0.5)

39.9 (± 1.1)

  1. The performance metrics of the multivariable models were evaluated using the test set by calculating the mean and standard deviation across the fivefold models. The specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy at a sensitivity threshold of no less than 90% were presented. The best results for each metric are indicated in bold.
  2. T-size Tumor size; MLP Multilayer perceptron; ROC Receiver operating characteristic; AUC Area under the curve; PR Precision recall; TPR True positive rate; TNR True negative rate.
Back to article page