Table 3 Comparison of the performances obtained for different models, in the scenario of balancing distributions.

From: Analyzing artificial intelligence systems for the prediction of atrial fibrillation from sinus-rhythm ECGs including demographics and feature visualization

 

AUC

Sensitivity

Specificity

DOR

F1 score

Resnet

0.77 (0.70–0.84)

65.00% (51.60–76.87)

70.67% (65.16–75.76)

4.47 (2.49–8.04)

0.42 (0.30–0.45)

Resnet with age/sex

0.79 (0.72–0.86)

78.33% (65.80–87.93)

69.33% (63.78–74.50)

8.17 (4.22–15.84)

0.47 (0.35–0.50)

MLR

0.71 (0.64–0.78)

66.67% (52.53–78.91)

67.32% (60.44–73.69)

4.12 (2.18–7.79)

0.46 (0.36–0.54)

  1. In the scenario of balancing distributions, we compare the three models considered. To compute the metrics additional to AUC, we set a threshold for each model, as described in “Methods”. Performances are assessed according to the aggregation rule by average. 95% confidence intervals are reported in brackets.
  2. M Male, F Female, AUC Area Under the Curve, DOC Diagnostic Odd Ratio.
Back to article page