Table 1 Feature extraction model & sPDA prediction model performance.

From: Extraction and evaluation of features of preterm patent ductus arteriosus in chest X-ray images using deep learning

 

Sensitivity

Specificity

ROC-AUC

CNX (Combined, Human-guided)

0.33

0.91

0.72

CNX (Combined, Rule-based)

0.02

0.99

0.40

CNX (Raw, Human-guided)

0.38

0.77

0.62

CNX (Heart, Human-guided)

0.22

0.90

0.63

CNX (Thorax, Human-guided)

0.49

0.70

0.61

XGB (Vector + Clinical)

0.33

0.95

0.75

XGB (Clinical)

0.29

0.95

0.67

XGB (Vector)

0.33

0.94

0.71

XGB (Ratio + Clinical)

0.42

0.94

0.74

XGB (CTR + Clinical)

0.40

0.93

0.73

  1. Sensitivity, specificity, and ROC-AUC of the ConvNeXt-based feature extraction model and the XGBoost-based sPDA prediction model. Combined, Raw, Heart, and Thorax refer to the input images used for training the feature extraction model (see Fig. 2). For the XGB model, Vector refers to the 1024 size feature vectors extracted from the feature extraction model, Clinical refers to the clinical features of gestational age, birth weight, and gender, and Ratio refers to the three ratio-based features that showed statistically significant differences between sPDA-positive and sPDA-negative groups (see Fig. 5). CTR, Cardiothoracic ratio.
Back to article page