Table 1 Comparison of AUC results achieved for different models, according to different aggregation rules.

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

Model

All ECGs–AUC

Avg ECGs–AUC

Max ECGs–AUC

Resnet

0.84 (0.78–0.90)

0.87 (0.81–0.93)

0.84 (0.78–0.91)

Resnet with age/sex

0.85 (0.80–0.91)

0.89 (0.83–0.95)

0.87 (0.81–0.93)

Resnet + Recurrent layers

0.83 (0.78–0.89)

0.87 (0.81–0.93)

0.84 (0.78–0.91)

RNN

0.77 (0.70–0.83)

0.81 (0.74–0.88)

0.75 (0.68–0.83)

Wavelet Scattering

0.78 (0.71–0.84)

0.80 (0.73–0.87)

0.77 (0.69–0.84)

MLR

0.85 (0.79–0.91)

0.86 (0.80–0.93)

0.85 (0.78–0.92)

Fusion

0.85 (0.79–0.91)

0.87 (0.81–0.94)

0.86 (0.79–0.92)

  1. The performance of six different models is evaluated with AUC. AUC for the Fusion model is computed by averaging for each ECG sample the six AF scores computed by the different models. The same data have been used to train and evaluate the six models. The only exception refers to MLR and Fusion models, where ECG samples with missing discrete values have been discarded from the analysis. AUC results are computed according to three different aggregation rules: i) no aggregation (All ECGs), ii) average of AF scores related to the same patient (Avg ECGs), and iii) selection of the highest AF score for each patient (Max ECGs). 95% confidence intervals are reported in brackets.
  2. ECG Electrocardiogram, AUC Area Under the Curve, RNN Recurrent Neural Network, MLR Multivariate Logistic Regression.
Back to article page